Infant pattern vision: A new approach based on the contrast sensitivity function

JOURNAL

OF EXPERIMENTAL

CHILD

PSYCHOLOGY

31, 1-45 (1981)

Infant Pattern Vision: A New Approach Based on the Contrast Sensitivity Function MARTIN

S. BANKS

The University of Texas at Austin AND

PHILIP

SALAPATEK

University of Minnesota Current approaches to the study of infant pattern vision have yielded interesting findings but have not yielded a set of data or principles from which general predictions can be drawn. We propose an alternative approach based on measurements of the contrast sensitivity function (CSF). This approach has been successfully applied to the study of adult vision. In principle, the approach allows one to predict the detectability of a wide variety of two-dimensional patterns if one knows the observer’s CSF. Two experiments were conducted. In Experiment 1, CSFs of I-, 2-, and 3-month infants were measured using a fixation preference paradigm. The results indicated noteworthy development between I and 3 months particularly in sensitivity to high spatial frequencies (fine stripes). The lowfrequency attenuation characteristic of adult vision is observed at 2 and 3 months but not always at I. In Experiment 2, CSFs of 2-month infants were measured at a lower luminance level. The results indicated that low-frequency attenuation became less pronounced as would be predicted if it were a manifestation of lateral inhibitory processing. The manner in which the CSF can be used to make general predictions is described. The CSFs of Experiment 1 are then used to successfully predict infants’ detection of patterns used in two frequently cited experiments. We also propose a simple model of infant pattern preference and show that the model accurately predicts the results of a number of well-known experiments. This research was supported by a University Research Institute summer research grant to the first author at the University of Texas. and by NIH Grant ROI-HD-07317 to the second author, and NlCHD Grants HD-01136 to the Institute of Child Development and HD-00098 to the Center for Research in Human Learning at the University of Minnesota. Parts of this paper were submitted to the Graduate School of the University of Minnesota in partial fulfillment of the requirements for the Doctor of Philosophy degree. We thank Velma Dobson, Wilson Geisler, Gordon Legge, Morton Mendelson, Douglas Sawin. and Davida Teller for comments on an earlier version. We also thank Michael Becker for assisting in the preparation of Fig. 14. Address reprint requests to Martin S. Banks. Department of Psychology, The University of Texas at Austin, Austin, TX 78712.

0022-0965/81/010001-45$02.00/O Copyright 0 1981 by Academic Press. Inc. All rights of reproduction in any form reserved.

2

BANKSANDSALAPATEK

The content of an inexperienced infant’s perceptual world has long been an object of speculation. Empirical evidence relevant to such speculation remained elusive. however, until experimental techniques wege refined for studying young infants’ perceptual capabilities. Stirnimann (l944), Fantz (1958), and Berlyne (195X) used the fixation preference (preferential looking) technique to demonstrate that young infants can respond to the presence of visual stimuli and even respond differentially to some pairs of visual stimuli. These pioneering studies thus indicated that young infants possess some perceptual capability and. consequently, that their perceptual world is not totally chaotic. The aspect of infant perception that has attracted the most experimental interest is the visual perception of pattern. The stimuli used in studies of infant pattern vision have generally been two-dimensional. achromatic. and stationary. So the implicit definition of pattern has been a nonrepresentational distribution of contours in a two-dimensional plane (Zusne. 1970). Since the appearance of the early preference studies, numerous researchers have shown that infants possess some ability to differentially respond to stimuli differing only in pattern. Given that the question of the presence or absence of some pattern vision capability has been answered, more recent research has turned to the problem of accurately characterizing this capability. Before discussing this research, however. definitions of three terms are needed: pattern detection. pattern discrimination. and pattern preference. Pattern detection refers to the ability to perceive the presence of a pattern rather than a uniform field. For example. in an adult detection experiment. subjects might be asked to respond “yes” whenever a small spot of light is perceived and “no” when it is not. Pattern discrimination refers to the ability to distinguish two (or more) detectable patterns from one another. In an adult discrimination experiment, subjects might be asked to respond “same” when two spots of light appear to be identical and “different” when they do not. Pattern preference refers to consistent spontaneous fixation of one pattern rather than another once the pattern(s) have been detected and/or discriminated. A large number of experiments have investigated early pattern vision in the last two decades. For example, fixation preference studies have shown that infants exhibit differential preferences for stimuli differing in linearity (Fantz. 1965; FdntZ & Nevis. 1967; Fantz. Fagan. & Miranda. 1975). complexity (Brennan, Ames, & Moore. 1966; Greenberg & O’Donnell. 1972; Hershenson, 1964; Hershenson. Munsinger, bt Kessen. 1965. and others). contour density (Karmel, 1969a, 1969b: Karmel & Maisel. 1973, number of orientations (Ruff & Birch, 1974), size (Fantz, Ordy, & Udelf. 1962: Fantz rf trl., 1975). regularity (Fantz. 1965: Karmel, 1969b: McCall & Melson, 1970). symmetry (Fantzrr ~11.. 1975). and concentricity (Ruff & Birch. 1974). There is also a growing literature on the young infants‘ ability to simply detect the presence of visual patterns. For

INFANT

PATTERN

3

VISION

example, several investigators have measured infants’ ability to detect pattern as a function of the size of elements within the pattern. A variety of stimuli have been used for this purpose: equally spaced, light and dark stripes (square wave gratings), checkerboards (reviewed by Dobson & Teller, 1978), unequally spaced, light and dark stripes (rectangular wave gratings), concentric light and dark rings (Fantz et uf., 1975). single vertical bars (Lewis, Maurer, & Kay. 1978) and others. Thus. the infant literature provides a rather lengthy list of stimulus dimensions along which infants can discriminate and preferentially fixate visual patterns. Similarly, the literature concerning how various stimulus dimensions. such as element size, affect the detectability of patterns is increasing. Nevertheless, these data (particularly the discrimination and preference data) contribute minimally toward a general understanding of infant pattern vision. The preference literature, for example, provides data concerning the relation of a number of stimulus dimensions to infant preferences. Despite this, the number of stimulus dimensions which are potentially relevant to the description of all patterns infants can process differentially is even larger, if not infinite.’ Unfortunately, no one has provided reasonable guidelines for generalizing findings obtained with well-studied patterns to other, as yet unstudied, patterns. If this problem is not solved, we will have to endlessly expand the list of dimensions whose influence on visual preferences has been studied (and also determine the relationships among those dimensions) to gain a general understanding of infant pattern preferences. A similar argument applies to the study of pattern detection in young infants. The current literature does not provide a general characterization because one cannot predict the detectability of one set of patterns from results obtained with another set of patterns. So, for example, measurements of the minimal detectable check size cannot be used presently to predict the minimal detectable diameter of a single disk on a background. A more effective approach to characterizing infant pattern vision would be possible if we could find a limited set of dimensions which could be used to describe all visual patterns and then could determine the relation’ The

following

example

illustrates

this

vance of two dimensions of pattern-number newbor-n fixation preferences. Both size and that is, when the value along one dimension other dimension led to changes in preferential the only relevant number of patterns varied the linearity.

pornt.

Fantz

et ul.

(1975)

investigated

the

and size of elements in a configuration-to number were shown to be relevant dimensions: was held constant, changes in value along looking. However, number and size are

stimulus dimensions of infant visual selectivity. There which have identical numbers and sizes of elements. regularity. or symmetry of the element configuration

patterns. differential preferences might reasonably be observed. contrast. orientation. or shape of the individual elements could preferences. One could presumably continue to add a considerable before enough had been included to allow a general description.

Similarly, also lead number

rele-

the not

are an infinite Certainly if one in two such changes in the to differential of dimensions

4

BANKS

AND

SALAPATEK

ship between those dimensions and infant perceptual responsivity. With such an approach adult psychophysicists have been able to use findings obtained with a limited set of patterns to make general predictions of how detectable other types of patterns are to adults (Cornsweet, 1970; Davidson, 1966; Ratliff, 1965). In this paper we follow this lead by describing such an approach and demonstrating how its use could in principle provide a more general characterization of pattern detection, discrimination. and preference in young infants. (We will argue that the approach is best suited to detection and preference, however.) This approach, which involves the measurement of the contrast sensitivity function (CSF). has only recently been applied to infant pattern vision. Therefore, we will describe the CSF approach. review the current infant CSF literature, present some detailed measurements of CSFs in young infants, and discuss the visual mechanisms which underlie developmental changes in those CSFs. In the discussion section of this paper, we will fully describe the theory behind the CSF approach and show how it allows one to predict the results of different infant experiments. We will also develop a model of infant pattern detection and preference and discuss its relation to other current models. The limited set of patterns which form the basis for the CSF approach are sine wave gratings. Two examples of sine wave gratings are shown in Fig. 1. Sine wave gratings are specified by four parameters: ( I) spatial frequency, the number of dark bars per degree of visual angle (spatial

Liii 2-5vvvvv POSITION

SPATIAL

FIG I, frequency

Two sine wave representations.

POSITION

FREQUENCY

gratings and In the upper

SPATIAL

the corresponding part of the figure,

FREQUENCY

intensity two sine

distributions wave gratings

and spatial are shown;

the grating on the right has a higher spatial frequency. Directly below them the corresponding intensity distributions are displayed: intensity is plotted as a function of position. The spatial frequency components of the gratings are shown in the lower part of the figure: gratmg contrast is plotted as a function of spatial frequency.

INFANT

PATTERN

VlSION

5

frequency is higher in the grating on the right in Fig. I), (2) orientation, the grating’s tilt to the left or right of vertical (the ones in Fig. I are vertically oriented), (3) phase, the grating’s position with respect to some reference position and, (4) contrast, which is related to the difference between the peak and trough intensities of the grating (formally contrast is defined by the equation: C = [I,,, - I,& / [I,,, + Zmin] where I,,, is the intensity of the most intense part of the light stripes andI,, is the intensity of the least intense part of the dark stripes). The CSF itself is determined by measuring an observer’s contrast sensitivity to sine wave gratings of various spatial frequencies. Typically an adult’s CSF is measured by presenting gratings of a number of different frequencies one at a time and determining the least contrast necessary for the adult to detect the grating at each of those frequencies. An example of a typical adult CSF is shown in the lower portion of Fig. 2. Contrast sensitivity, the reciprocal of the minimum contrast required for detection, is plotted as a function of spatial frequency. Note that sensitivity is greatest for intermediate spatial frequencies (2 to 6 cycles (cy)/deg) and lower for low and high frequencies. The upper part of the figure is a sine wave grating which increases in spatial frequency from left to right and increases in contrast from top to bottom. We have included this to give the reader a feeling for what the CSF represents. The physical contrast of the grating is constant along any horizontal line in the photograph, but itsperceived contrast varies notably with spatial frequency. Indeed, the highest perceived contrast is for intermediate frequencies. Note the correspondance between your ability to detect the grating at different frequencies in the upper part of the figure and the CSF plotted in the lower part. The theoretical utility of the CSF derives from Fourier’s theorem and linear systems analysis. Both Fourier’s theorem and systems analysis will be described fully in the discussion section. Suffice it to say here that Fourier’s theorem states that any two-dimensional, stationary stimulus can be exactly described by the combination of a set of sine wave gratings of various spatial frequencies, orientations, contrasts, and phases. Thus, even a complex, two-dimensional pattern such as the picture of a face can be exactly reproduced by the combination of various sine wave gratings. Thus, in an important sense the CSF characterizes the pattern information to which an observer is sensitive. Linear systems analysis capitalizes on this fact. In principle, it allows one to predict the visibility of any pattern if one knows the observer’s CSF. (Limitations on this predictive power will be described in the discussion section.) There are numerous examples in the adult literature where the perceived quality of a particular pattern was successfully predicted using the CSF and linear systems analysis. Campbell, Carpenter, and Levinson (1969), for example, accurately predicted the visibility of single lines of various widths and contrasts from their observers’ CSFs. Lowry and

BANKS

6

AND

SPATIAL I 20/2ccO

FREQUENCY

I

1

/IO00

1500

I

I

100

50

1 /zoo

Sine of the

wave figure

grating displays

(CY/DEQ)

1

I

I

/IO0

150

120

SNELLEN NOTATION I I 1 20

IO

MINUTES FIG. -.’ pper part

SALAPATEK

and typical a sine wave

5

OF

I 2OAO

1

I

I

2

1

5

ARC

adult contrast grating whose

sensitivity function spatial frequency

(CSF). increase5

The from

INFANT

PATTERN

VISION

7

DePalma (1961) and others have shown that adults’ perception of “Mach bands” near light-dark edges can be predicted from the CSF and systems analysis. Kelly and Magnuski (1975) demonstrated that the detectability of concentric circular patterns can be similarly predicted. In addition to providing a characterization of the pattern vision capabilities of an observer. the CSF reflects some basic properties which might underlie development in pattern vision. Visual acuity, for example, is often defined as the highest frequency grating an observer can detect. Under optimal conditions this frequency is about 50 cyideg in adults (see Fig. 2). Another important property is sensitivity to contrast. The contrast sensitivity values represented by the CSF provide indices of how well an observer can discern small differences in intensity. Yet another important property is low-frequency attenuation. Adults’ sensitivity to low spatial frequencies is lower than it is to intermediate frequencies. Hence, the adult CSF exhibits a low-frequency falloff in sensitivity (see Fig. 2); the functional significance of this has been discussed by several authors (e.g.. Cornsweet, 1970; Ratliff, 1965) and will be mentioned in the discussion section of this paper. Three groups of investigators have recently measured the CSFs of infants at different ages (Atkinson, Braddick, & French, 1979: Atkinson, Braddick. & Moar, 1977a. 1977b; Banks & Salapatek, 1978; Pirchio. Spinelli. Fiorentini. & Maffei, 1978). Atkinson and her colleagues (1977a. 1977b) measured CSFs in I-. I!-, and 3-month-old infants using a preferential looking paradigm (Peeples & Teller, 1975). Two stimuli-a sine wave grating (that was either stationary or drifting) and a uniform field of equal average luminance-were presented on each trial. one to the left of center and one to the right. The stimuli were relatively small (15” circular fields) and were separated by a 9” space that contained central fixation lights. The response measure was a “blind” adult observer’s best guess. based on the infant’s behavior, of the side the grating had been on. Generally, three or four spatial frequencies were presented to each infant: contrast sensitivity (the reciprocal of contrast at threshold) for each frequency was estimated by finding the contrast value associated with 70% correct responding. Pirchioet al. (197X) used the visually evoked response (VER) to measure CSFs in 2 f, 3 f, and 6-month-olds. The stimuli were flickering (phase-alternated) sine wave gratings of various spatial frequencies. The stimuli in this study were also relatively small. varying from 7” to 2.5” for

left to right and contrast increases from top to bottom. The lower part of the figure shows a typical adult CSF. Contrast sensitivity, the reciprocal of contrast at threshold, is plotted as a function of spatial frequency. Scales relating spatial frequency to Snellen equivalents and stripe width in minutes of arc are shown for comparison. If the figure is viewed from a distance of 2.5 m. the scales at the bottom indicate the actual frequency values of the grating in the upper part of the figure. (Adapted from Comsweet. 1970.)

8

BANKS

AND

SALAPATEK

various conditions. Contrast thresholds at each frequency were determined by a normalization procedure. Atkinson et ul. (1977a, 1977b) and Pirchio et al. (1978) found young infants’ contrast sensitivity and acuity to be markedly reduced relative to adults’. They also observed clear age-related changes: the low-frequency falloff became more marked with age (see particularly Atkinson et al., 1977b) and both sensitivity to contrast and visual acuity increased. We measured CSFs in I-, 2-, and 3-month-olds using a fixation preference procedure. We used larger stimuli than Atkinson et al. and Pirchio et al. in the hope of minimizing distraction due to the external contour of the gratings (Milewski, 1976). We also used stationary, nonflickering gratings for stimuli and first fixation as the response measure to allow a potentially more accurate estimation of the CSF’s low-frequency falloff in young infants.? In the following sections, we provide a detailed description of the methodology and results of our earlier brief report (Banks & Salapatek. 1978). These data provide information concerning the development of visual acuity, low-frequency attenuation, and sensitivity to contrast. We also present the results of a second experiment specifically concerned with the development of low-frequency attenuation. The results of these experiments provide an approximate picture of the pattern information available to I-, 2-, and 3-month-old infants and, as we will show in the discussion section, can be used to predict detection of and. perhaps, preference for a relatively wide variety of patterns. EXPERIMENT

1

Methods

The apparatus is schematized in Fig. 3. Sine wave gratings were produced optically in a manner similar to that described by Davidson (1966). Transparencies (T) were prepared with a sinusoidal profile over half of their horizontal extent and a straight line profile over the other half. The transparencies were opaque below the profile and transparent above it. A given transparency was imaged at the plane of the flashed opal glass stimulus field (F) by the objective lens (L,.J, front-surface mirror (M,), and a four-sided rotating mirror (M,). The rotating mirror served to sweep the image vertically across the stimulus field at a rate of 120 Hz. In this * Adult

psychophysical

research

has

shown

that

contrast

sensitivity

to low-frequency

gratings is greater for flickering or drifting gratings than for stationary, nonflickeringgratings or when the subject scans a stationary, nonflickering grating rather than maintaining steady fixation (e.g., Robson, 1966). This is presumably due to the fact that the low-frequency attentuation characteristic of lateral inhibition is weakened by temporal modulation of the light distribution falling on various retinal loci. We designed our procedure to minimize such temporal effects in order to best estimate the development of low-frequency attentuation. By using fixate

first one

fixation pattern

as the response over another

measure. was made

we hoped to ensure before any significant

that

the infants’ scanning had

decision occurred.

to

INFANT

LS

L,

T

L,

PATTERN

S

9

VISION

M,

a-?T[q

FIG.

3.

Schematic

of the

experimental

apparatus.

quality, achromatic lenses: T, transparency; M,. four-sided rotating mirror; A. aperture; transparency (from Banks

is shown in the lower & Salapatek, 1976).

left

S. shutter, F, flashed and

the

LS,

light

M, opal

and M,, front-surface glass stimulus field.

corresponding

source;

stimulus

L,

and

in the

b,

high-

mirrors; A typical lower

right

manner a sine wave grating was produced across one half of the field and an immediately adjacent uniform field across the other half. The adjacent positioning of the grating and uniform half-fields was intended to maximize the probability of infants reponding on the basis of fovea1 processing. The four sides of the rotating mirror (M,) were within 0.005 in. of true parallel about the axis of rotation. Furthermore, the orientation of T, M,, and the axis of M, were finely adjusted so no significant reduction in contrast within the range of spatial frequencies used (0. I5 to 2 cy/deg) was observed. The contrast of each stimulus used in the experiment was carefully measured photometrically and that measured contrast was used in all analyses. The range of contrasts that could be presented varied from .70 to .OOl. Contrast and spatial frequency were varied by inserting different transparencies in the apparatus. Contrast varied with the peak-to-trough amplitude of the sine wave portion of the transparencies; spatial frequency varied with the number of cycles. The space-average luminance of the grating half-field and the uniform half-field was 55 cd/m’ (16 ft long). The gratings were always vertical in orientation. The stimulus field was very large, 96“ x 40” (118 x 49 cm), to maximize the infants’ attention to it and to ensure that a sufficient number of cycles was present in the low-frequency gratings. The field was reflected by a large front-surface mirror (M3) so that infants could view the stimuli while lying in a supine position. The optical distance between the infants and the stimulus field was 53 cm. This distance was chosen because acuity estimates in l- and 2-month-olds are as high at this viewing distance as at any other (Atkinson et al., 1977b; Fantz, et al., 1962; Salapatek, Bechtold, & Bushnell, 1976). Furthermore, recent measurements of accommodation and depth of focus in I-, 2-, and 3-month-olds suggest that no noticeable

10

BANKS

AND

SALAPATEK

loss of contrast sensitivity should be observed at this viewing distance (Banks, 1980). A second part of the apparatus (not shown in Fig. 3) imaged a vertical bar at the midline of the stimulus field between presentations of the, gratings. When an observer judged that the infant was fixating the bar, a 3-set presentation of the grating and uniform half-fields ensued. Gratings of variable contrast and spatial frequency were presented to either the left or right of center. Two observers, who could not see the stimulus field, indicated whether the infants’ first fixation was to the left or right. The observers’ responses were automatically recorded on an event recorder. The experimenter determined the number of hits (correct responses) and misses (incorrect responses) on-line by comparing the observers’ responses to the side on which the grating had been presented. A hit was recorded when both observers indicated that the first fixation was to the side the grating had been on. A miss was recorded when both indicated that the first fixation was to the side the uniform field had been on. When the observers’ responses were contradictory (one indicated the first fixation was to the left and the other indicated it was to the right), the experimenter used a random array to assign a hit or miss to the trial. When the observers did notindicate that an eye movement had occurred, a similar random procedure was used. This did not occur very often since the observers were encouraged to guess when they were uncertain of the direction of first fixation. The observers agreed on the direction of first fixation on 8% of the trials (including those in which they presumably guessed). There were no significant differences in interobserver agreement for the three age groups. A descending and ascending staircase procedure was used to tind contrast thresholds at each of five spatial frequencies: 0.1. 0.3. 0.5. I .O, and 2.0 cy/deg. These spatial frequencies correspond to stripe widths of3, I .5, 0.9. 0.4. and 0.2 cm at the 53-cm viewing distance. The procedure continued at a given spatial frequency until a contrast associated with a hit rate below 75% and a contrast associated with a hit rate above 75%’ had been presented.:’ A total of 20 trials at each of these contrasts was 3 The according

staircase to the

procedure following

by rules.

which contrast A contrast value

thresholds was first

were chosen

determined which was

proceeded likely to be

suprathreshold. Ten trials at the contrast level were then presented. If performance was 75% or greater (eight or more “hits”). the contrast value was reduced by 50% and IO more trials were presented. If performance at the lower contrast still exceeded 75% correct responding. contrast was agatn reduced by 56%. This descending staircase continued until performance in IO trials dropped below 75% in which case IO additional trials were presented for a total of 20 at that contrast value. At this point contrast was increased and an ascending series of IO trials at given contrast performance over 20 exceeded 75% correct.

values was presented until trials (IO from the descending

a contrast level and IO from

was found the ascending

at which serie\)

INFANT

PATTERN

VISION

11

presented. The final estimate of contrast thresholds was obtained by interpolation to the contrast associated with a hit rate of exactly 75%. A total of 200 to 300 trials was presented to each infant. Data were collected on 11 l-month-old infants who ranged in age from 29 to 40 days. Complete CSFs were obtained for six of these infants. We tested 14 2-month-olds ranging in age from 59 to 71 days. Complete CSFs were obtained for six of these infants. The low-frequency slope of the CSF was determined for four additional infants in this age group. Data were collected on 10 3-month-olds who ranged in age from 87 to 101 days. Complete CSFs were obtained for eight of these infants. Measurement of a CSF for one infant generally required two 30- to 45min sessions. Infants for whom complete CSFs were not determined were terminated due to fussiness. sleepiness, or failure to return for a second session. The same apparatus was used to measure an adult CSF, but the procedure was somewhat different. The adult sat upright 53 cm from the stimulus field and fixated a small fixation point during the entire experimental session. This subject initiated trials himself and indicated by button presses whether he thought the sine wave grating had been presented to the left or right of center. The subject was instructed not to make eye movements but to maintain fixation on the fixation point. The procedure for obtaining the adult’s contrast thresholds was identical to the procedure used with the infants. Two hit rates, one above 75% and one below, were obtained and linear interpolation between these two points yielded the contrast level associated with a hit rate of exactly 75%.

The CSF for the adult observer is shown in Fig. 4 (the average Z-month CSF is also shown). Contrast sensitivity, the reciprocal of the contrast associated with a 75% hit rate, is plotted as a function of spatial frequency. Note that the low-frequency falloff typical of adult CSFs is observed and that overall contrast sensitivity is quite high. The CSFs obtained for six I-month-olds are shown in Fig. 5. Contrast sensitivity is plotted as a function of spatial frequency. There are several things to note about these functions. First, sensitivity to spatial frequencies higher than 1 cy/deg was generally very low. Second, four of the six l-month-old functions did not exhibit the low-frequency falloff typical of adult CSFs. In other words, those four infants were just as sensitive to low spatial frequencies as they were to intermediate frequencies. However, two other infants, J.M. and L.G., did show the low-frequency falloff. Third, the infants’ contrast sensitivity was generally much lower than the adult’s. With the most sensitive infant, a contrast of about .03 (contrast sensitivity = 33) was necessary to elicit 75% correct responding at the best spatial frequency. The adult observer elicited similar performance at a contrast of only .0017 (contrast sensitivity = 590).

12

BANKS

AND

SALAPATEK

FIG. 4. Adult and infant CSFs. Both CSFs represent contrast sensitivity. the reciprocal of contrast at threshold, as a function of spatial frequency. In the adult CSF the open symbols and solid fines represent data obtained with the experimental apparatus. Each symbol represents an interpolated value (see text for explanation). The broken line is an idealization of a typical adult CSF under similar stimulus conditions (Schade. 1956). The (see Fig. 6). infant CSF represents the average values obtained for 2-month-olds

Figure 6 shows the CSFs obtained for the six 2-month-olds. Once again sensitivity to high spatial frequencies was notably low. All six of the 2-month-olds exhibited a low-frequency falloff in sensitivity. To check the generality of this finding. four additional 2-month-olds were tested on the three lowest spatial frequencies. All four also exhibited a low-frequency falloff, although the falloff was quite small for one of them. Peak contrast sensitivity was once more quite low in comparison to adult sensitivity. I MONTH

OLDS (16ftL

SpatialFrequency FIG. function

5.

J

lcyldegi

The individual CSFs for six I-month-olds. Contrast sensitivity of spatial frequency. Average luminance was 55 cd/m2.

is plotted

as i

INFANT

PATTERN 2 MONTH

FIG.

6.

The

individual

CSFs

OLDS

13

VISION (16 ft L )

for six Z-month-olds

(luminance

= 55 cd/m2)

Figure 7 shows the CSFs obtained for the eight 3-month-olds. Sensitivity to high spatial frequencies was still quite low, but notably higher than in the l- and 2-month CSFs. All of these infants exhibited a low-frequency falloff with the exception of R.S. Maximum contrast sensitivity was higher in the I- and 2-month-olds but still notably low compared to adult sensitivity. An analysis of variance was conducted on these data with age as a between-subjects factor and spatial frequency as a within-subjects factor.

SPoflal Frequency (cy/derg)

FIG.

7.

The

individual

CSFs

for eight

3-month-olds

(luminance

= 55 cd/m*)

14

BANKS

AND

SALAPATEK

The age main effect was significant (F( 2, 17) = 5.85.~ = .012). The spatial frequency factor was also significant (F(4, 8) = 9.62, p = ,004) as was the age x spatial frequency interaction (F(8. 68) = 3.20, p = .004). Figure 8 shows the average CSFs for each age group: these functions illustrate the age-related changes in the CSF. As mentioned earlier, the CSF provides an accurate and reliable estimate of visual acuity. One can see from Fig. 8 that sensitivity to high spatial frequencies increased notably between 2 and 3 months. Campbell and Green (1965) and others have shown that adult acuity can be estimated by replotting CSFs on a linear spatial frequency scale (rather than a logarithmic scale as in Fig. 8) and then linearly extrapolating the highfrequency slopes of these functions to find the spatial frequency associated with a contrast of 1.O. This spatial frequency is an estimate of the finest resolvable high-contrast grating. We used the same technique to estimate infants’ acuity. Straight lines were fit using a least-squares criterion through the points at 0.5, 1. and 2 cyideg for each of the infants in each of the age groups. The spatial frequencies cut by these lines at a contrast of 1.0 are 2.4 cyideg (SE = .4) for the I-month-olds, 2.8 cy/deg (SE = .4) for the 2-month-olds, and 4.0 cyideg (SE = .8) for the 3-montholds. In Snellen notation. these correspond to acuities of 20/250 for the I-month-olds. 201215 for the 2-month olds, and 201150 for the 3-montholds. Statistical tests performed on these “cut-off” spatial frequencies indicated that 3-month-old values were significantly greater than l-month-old values (t = 4.3 I, p < .()OZ, one-tailed) and 2-month-old values (t = 3.3 1. p < ,005, one-tailed). The difference between the I- and 2-month values approached significance (t = 1.44. p < .lO. one-tailed). Age differences in the low-frequency portion of the CSFs were also observed. The average CSF for I-month-olds exhibited no low-frequency falloff. In contrast, the 2- and 3-month functions exhibited marked falloffs. Note that low-frequency insensitivity was observed at 0. I5 and

, i,-..-1-4 2 Spatial

FIG. contrast

8.

The average CSFs sensitivity is plotted

.5

I

Frequency

for I-. 7-, L and as a function

2 (cy/deg)

3.month-olds of spatial

(luminance frequency.

= 55 cd/m’).

Average

INFANTPATTERN

VISION

1.5

0.3 cyideg among the 3-month-olds but only at 0. I5 cyideg among the 2-month-olds. To perform statistical tests on the low-frequency portion of the CSFs, lines were fit (using a least-squares criterion) through each subject’s data at the lower two or three spatial frequencies. The slopes of these lines were then calculated. The mean slopes (log contrast sensitivity as a function of linear spatial frequency) were - .34. 3.46, and 1.15 for I-, 2-, and 3-month-olds, respectively. The slopes for the 2-month-olds were significantly greater than zero (t = 4.43, p < .005, one-tailed) as were the slopes for the 3-month-olds (t = 3.29. p < .025, one-tailed). The I-month-olds’ slopes were not significantly greater than zero (r = 0.26, n.s.). Thus our data suggest that the low-frequency falloff is generally present in 2- and 3-month-olds and that it is generally not present in I-month-olds at least for the range of spatial frequencies we tested. EXPERIMENT

2

The low-frequency falloff in CSFs is generally believed to be a manifestation of lateral inhibitory processing in the visual system. Lateral inhibitory processing refers to inhibitory interactions among neighboring neural elements and is commonplace in the vertebrate retina (Cornsweet, 1970; Ratliff, 1965; see discussion section for more information). The evidence for a low-frequency falloff in the 2- and 3-month-olds does not necessarily mean, however, that they are insensitive to low spatial frequencies nor that the falloff is due to lateral inhibitory processing. It could reasonably be argued that 2- and 3-month-olds can detect low frequencies but may simply not prefer to fixate them as consistently as they do intermediate frequencies. One way to attack this problem is to predict how the low-frequency falloff should change under various stimulus conditions if it does in fact reflect lateral inhibitory processing. It is known from physiological studies that the magnitude of lateral inhibition in the retina is less for low luminances than for high luminances (Barlow, Fitzhugh, & Kuffler, 1957; Enroth-Cugell & Robson, 1966). Human adult psychophysical studies also indicate that the slope of the CSF’s lowfrequency falloff is positively correlated with stimulus luminance (Van Nes & Bouman. 1965). Therefore, several investigators (e.g., Kelly. 1975) have argued that the CSF’s low-frequency falloff reflects lateral inhibitory processing and. furthermore, that changes in the falloff with changes in stimulus luminance reflect changes in the magnitude of lateral inhibition. Given this, we measured I?-month-olds CSFs at a reduced luminance of 9.2 cd/m2 to see if a predictable decrease in the steepness of the lowfrequency falloff would be observed.

The apparatus and procedure were identical to that of Experiment I. The space-average luminance of the grating and uniform half-fields were reduced by .77 log units to 9.2 cd/m” (2.7 ft long).

16

BANKS

AND

SALAPATEK

Data were collected on eight 2-month-old infants who ranged in age from 53 to 68 days. Complete CSFs were obtained for five of these infants. One adult was also tested in the same manner as in Experiment I. Results

The CSF obtained for the adult is shown in Fig. 9. Note that the low-frequency falloff is less pronounced than the one observed at the higher luminance (55 cd/m2) of Experiment I (see Fig. 4). This decrease in the falloff’s slope is similar to those observed in other adult psychophysical studies (e.g., Van Nes & Bouman, 196.5). Figure 10 displays the individual CSFs obtained for the five infants. Comparison of Figs. 6 and IO reveals that the lower luminance CSFs exhibited consistently lower slopes in the low-frequency falloff. To illustrate this, the average 2-month CSFs for Experiments 1 and 2 are shown in the lower right panel of Fig. 10. To perform statistical tests on these data, lines were again fit (using a least-squares criterion) through each subject’s data at the lower two or three spatial frequencies. The slopes of these lines were calculated and compared statistically to the 2-month slopes for 55 cd/m2 (Experiment 1). The slopes in Experiment I were significantly greater than the slopes in Experiment 2 (t = 2.18, p < .05. one-tailed). McCarvill and Karmel (1976) have recently reported somewhat similar results. The finding that the slope of the low-frequency falloff changed in a predictable manner strengthens the conclusion that the falloff in the ‘-month CSF is not due to a simple lack of preference to fixate low spatial frequencies and, furthermore, that the falloff is a manifestation of lateral inhibition.

FIG. ft long).

9. CSF Contrast

and solid lines an idealization

obtained sensitivity

with

the adult subject with an average luminance is plotted as a function of spatial frequency.

represent data obtained of a typical adult CSF

from under

the experimental similar stimulus

apparatus. conditions

of9.2 cd/m2 (2.7 The open symbols The broken (Schade.

line 1956).

is

INFANT

JC 1~ CR

FIG. cd/m’

IO. The individual (2.7 ft long). In the

by the solid symbols; the open symbols.

PATTERN

2 MONTH OLDS l27ftL) ~~ r 1‘8

CSFs for five 2-month-olds lower right. the average ?-month

the average

17

VISION

2-month

CSF

(see

Figure

with

an average

CSF 8) for

for

luminance

9.2 cd/m2

55 cd/m”

of 9.2

is represented

is represented

by

DISCUSSION The CSFs reported here provide an estimate of the pattern information available to I-, 2-, and 3-month-old infants. Visual acuity. sensitivity to contrast, and low-frequency attenuation are important aspects of the CSF which constrain the information which passes through the visual system. We will discuss each of these aspects of pattern vision in turn. We will then fully describe the theory behind the CSF approach and introduce some preliminary work on application of the CSF to predict the detectability of various patterns. Finally, we will discuss infant pattern preferences and describe a simple model, based on the CSF approach, which can successfully predict preferences to a variety of patterns. Visuul Acrrit) A basic property of pattern vision reflected in the CSF is visual acuity. Indeed, one common definition of acuity is the highest spatial frequency an observer can detect (Westheimer, 1972). The CSF index of adult acuity has been valuable because it can be used to predict the visibility of many types of acuity targets. For example, Campbell et al. (1969) have shown that adults’ acuity for a single line (minimum visible acuity) can be predicted from their CSFs. Our data (Experiment I) indicate that young infants can detect highcontrast gratings of 2.4, 2.8. and 4.0 cyideg at 1, 2, and 3 months, respectively. There are several other reports in the developmental literature concerning infant visual acuity (Atkinson, Braddick, & Braddick. 1974: Atkinson, et al., 1977b: Dayton, Jones, Aiu, Rawson, Steele &

18

BANKS

AND

SALAPATEK

Rose, 1964; Dobson & Teller, 1978; Fantz rt trl., 1962; Gorman, Cogan, & Gellis, 1957, 1959; Gwiazda, Brill, Mohindra, & Held, 1978; Harter & Suitt. 1970: Marg, Freeman, Peltzman. & Goldstein, 1976; Miranda, 1970; Pirchio et (11.. 1978: Salapatek et (il., 1976: Sokol, 1978; Sokol & Dobson, 1976: Teller, Morse. Borton, & Regal, 1974). Unfortunately. the acuity estimates reported in these studies are not always in close agreement. Figure 1 I displays various estimates obtained with sine wave gratings, square wave gratings, and checkerboards. This figure is similar to ones published by Dobson and Teller (1978). The variation in estimates from study to study may be due largely to differences in experimental technique (Dobson & Teller, 1978). Figure I I includes estimates obtained using optokinetic nystagmus (OKN) (Dayton et rrl.. 1964: Fantz cr al.. 1962; Gorman rt ~11.. 19.59). visually evoked cortical potentials (VEP) (Harter & Suitt. 1970, Marg et trl., 1976: Sokol. 1978), and preferential looking (Allen. 197X; Atkinson et al., 1977b: Fantz et trl.. 1962, 1975: Gwiazda et al., 1978: Miranda, 1970: Salapatek et al.. 1976; Teller et ~11.. 1974). Comparison of the results obtained using the three response measures indicates that the VEP studies generally obtained higher acuity estimates than either the OKN or preferential looking studies. This is most apparent at 6 months. Dobson and Teller (1978) argue that this discrepancy may be attributable to differences in the criteria for what constitutes threshold. The criteria in VEP acuity studies

FIG. II. summarized.

Visual acuity The average

age. Legend at the symbols and lines

as a function acuity estimate

right indicates represent the

from results

of age. reported

The results of several infant studies by each study is plotted as a function

which study each from experiments

data

point which

was used

paradigm. Open symbols and broken lines represent the results from used optokinetic nystagmus (OKN) or visually evoked responses (VER). the studies listed in the legend are: Allen (1978). Atkinson er ul. t 1977b). Fantz et cl/. t 1975). Miranda (1970). Salapatek ~1 trl. t 1976). Teller er rrl. Dayton et crl. t 1964). Gorman et trl. (1959). Harter and Suitt (1970). and Sokol (1978).

are of

obtained. Solid the preference

experiments which The references for Fantz et ul. t 1962). t 1974) t I. I cd/m’). Marg ef ul. (1976).

INFANT

PATTERN

VISION

19

have generally been rather lax; the criteria in OKN and preferential looking studies have generally been stricter.l For example. 75% correct responding was used as the threshold criterion in the present study. If a more lenient criterion had been adopted acuity estimates would have been correspondingly higher. Despite some discrepancies, the studies illustrated in Fig. I I reveal that acuity develops significantly during early infancy, but that infant acuity is quite poor in comparison to adult (until at least 5 to 7 months of age).” Thus, the ability to resolve detailed pattern information should improve with age. It should be emphasized, however, that knowing an infant’s acuity, 4 Recently estimates

Atkinson, obtained

Braddick. using

(VER). They found in threshold criteria

the

paradigm patterned

hypothesis

that

the

absence

seem There

to adequately are several

tion.

chromatic

factor quality

might (distances distances

to (ones

aberration,

that

young

in the

stimuli. why

visually

infant

is infant

diffraction to be large & Banks,

(Dobson which

response differences that

the

ability with

to the

paradigm

so poor

due enough 1978).

to the

pupil,

and

is ah-

how

does

mechanisms can be proposed that the optics of the infant’s

& Teller. affect adult

1978; Salapatek visual acuity:

and

in young infants Accommodative

clarity

of the

media however,

mature does

& Banks. spherical optic

infant

acuity

would

be

quite

and not

1978). aberra-

media.

to account for the low error. on the other hand.

imply

that

the

quality

of

poor

for

particular

None

observed might be

target

and would be Recent experiments, across Thu%. the

target distances accommodative

retinal

image

measured acuity in young infants. Thus. neural and/or attentional primary limit on resolution. Attentional factors cannot be ruled out that visually evoked potential and optokinetic nystagmus (OKN) agree relatively well with preferential looking measures (once criteria are considered) suggests that attention is not a primary There is considerable evidence that several neural factors are early acuity development. These factors include changes in the

to one

distance5

better for however, (Atkinson error

exceeds

factors at this stage, measures of differences limiting factor. significantly packing density

it to eye

Young infants may be unable to accommodate appropriately (Banks, 1980: Haynes, White, & Held. 1965). Therefore.

which they are not well accommodated) to which they are well accommodated).

considerations

acuity

acuity

evoked

preference

and that as the lens, cornea, and other optic acuity improves as well. This hypothesis.

that young infants’ acuity does not vary significantly 1977b: Fantz et a/.. 1962: Salapatek c’f ml.. 1976). appear to be a primary limitation to infant acuity. These

compared

the

similar acuity estimates once is consistent with the hypothesis

responding

two questions:

account for the data sorts of optical errors

viable candidate. at various distances expect

directly and

the first months of life’? Several Perhaps the simplest is to propose

for acuity improves.

of these errors is likely acuity values (Salapatek a more targets

have

paradigm

yielded finding

of differential

to distinguish motivate two

so rapidly over these questions. limiting image

(197’))

can provide reasonable estimates of the limits of infants’ from unpatterned stimuli. In other words. it is consistent

sociated with an inability s These observations

is the retinal

French preference

that the two techniques were minimized. This

preference distinguish

improve answer

and

fixation

does

other \how rt t/l.. not

behaviorall) must be the but the fact infant acuity in threshold involved in of fovea1

cones. migration of nonrecipient structures such as retinal ganglion cells away from the fovea1 pit. and spatial tuning of various visual neurons in the retina and central nervou\ system. For a more complete discussion of these factors. see Salapatek and Banks (197X). Regardless of the particular anatomical and physiological factors which determine the early growth of acuity. the primary resolution limitation does not appear to lie in the quality of the retinal

image

but

rather

in the

infant’s

ability

to proces\

that

image.

20

BANKSANDSALAPATEK

expressed in terms of cycles per degree. minutes of arc, or Snellen equivalents. is not sufficient to allow one to predict the detectability of a variety of acuity targets. Many types of visual patterns. such as single lines and Snellen letters, have been used to measure visual acuity in adults. It can be shown that an adult’s detection or identification of those fine visual patterns is dependent on contrast sensitivity for intermediate as well as high spatial frequencies. Because the CSF reflects sensitivity to a broad range of spatial frequencies, it allows one to predict the detectability of such acuity targets (e.g.. Campbell et al., 1969) and, thus, provides a useful general index of visual acuity.

Contrast is defined as an intensity ratio. The ability of the visual system to detect contrast is intimately involved in the perception of visual pattern, since intensity differences generally define patterns in the first place. Understandably the adult visual system is extremely sensitive to contrast. Increment thresholds (the increment of light necessary to detect a spot on a background) are as low as 3 to IO% (Brown & Mueller, 1965) and the contrast of a sine wave grating at threshold can be as low as 0.2% (Van Meeteren. 1967). There are few studies in the literature concerning infant sensitivity to contrast. Doris and Cooper (1966) attempted to measure young infants’ increment threshold for bright moving stripes on a darker background. Unfortunately, intensity and contrast were confounded in this experiment making interpretation difficult. Doris, Casper, and Poresky (1967) improved the original increment threshold experiment by partially untangling the confound between intensity and contrast. They found that newborns could detect intensity differences of 50% (change in intensity divided by background intensity) and I-month-olds. differences as low as 26%. Peeples and Teller (1975) also measured increment thresholds in young infants. In the initial phase of their experiment. four stationary vertical stripes differing slightly in intensity from the background were presented to the left or right of center. The infants consistently fixated the stripes when the intensity difference was only 10%. This corresponds to a contrast sensitivity of about 19. Detection of a single bar occurred with an intensity difference of about 25%; corresponding to a contrast sensitivity of about 8. Atkinson et al. (tY77b) measured the CSFs of I-. 2-. and 3-month-olds using stationary and drifting gratings. They found average peak contrast sensitivities of about 2.4, 5.5, and 10.4 at I. 2. and 3 months, respectively. Pirchio et ul. (1978) observed somewhat lower peak sensitivities: 4 and 8 at 2 3 and 3 4 months. The present study (Experiment I) yielded average peak sensitivities of IO. 12.5, and 15 for the three ages. Thus. sensitivity to contrast appears to increase during the first 3 months of life. The increase

INFANT

PATTERN

21

VISION

presumably continues well beyond that age range since 3-month-old sensitivity to contrast is still about 30 times lower than adult. The data of Pirchio et al. suggest, in fact, that sensitivity to contrast continues to increase until at least 8 months of age. It is somewhat inappropriate, however, to describe the contrast sensitivity of a visual system with a single value, since sensitivity to contrast is highly dependent on stimulus size and shape (Cornsweet, 1970: Ratliff, 1%5). Indeed, the interaction between contrast sensitivity and the spatial parameters of stimulation is a crucial aspect of pattern vision. Adult and infant visual systems are maximally sensitive to certain aspects of pattern information and insensitive to others. The definition of regions of relative sensitivity and insensitivity is equivalent to describing the aspects of pattern information adults and infants are tuned to. Thus, the contrast sensitivity values at a number of spatial frequencies are more useful than peak contrast sensitivity values which are useful only as a gross index of sensitivity. Low-Frequency

Attenuation

As mentioned before, the adult visual system is not very sensitive to low spatial frequencies. Low-frequency insensitivity is often assumed to be due to lateral inhibitory processing. This assumption has recently been questioned (Hoekstra, Van den Brink, & Bilsen, 1974; Savoy, & McCann, 1975) but more recent evidence (Estevez & Cavonius, 1976) indicates that the assumption is valid after all.6 Only a few studies including our Experiments 1 and 2 have investigated the low-frequency sensitivity of the infant visual system. Atkinson et al. (1977b) measured the CSFs of I-, 2-. and 3-month-olds and found evidence for a low-frequency falloff at 2 and 3 months but not at I month. Pirchio et al. (1978) measured the CSFs of 2 f-, 3 &. and 6-month-olds and observed low-frequency falloffs at all three ’ The stimulus inversely number argued limited present) of lateral

critics (Hoekstra field size is fixed proportional of cycles

et al., in most

1974; Savoy CSF studies

& McCann, so the number

to the spatial frequency of the grating. or bars of low-frequency gratings is typically

1975) have pointed out of bars (cycles) presented This

means presented.

that the low-frequency falloff observed in most CSF studies number of cycles in lower-frequency gratings (fewer intensity and, therefore, inhibition.

the falloff

might

be an experimental

artifact

that is

that only a limited Given this, they might be due to this gradient samples are

and

not

a manifestation

The adult data of Fig. 4 are relevant to this criticism. The CSF shown was measured with a very large stimulus field (48” x 40”) so that a sufficient number of cycles was present for the lower-frequency gratings. A low-frequency falloff was still observed. Thus, the lowfrequency falloff can still be observed when a sufficient number of cycles is present in the stimulus. Estevez and Cavonius (1976) have looked and Hoekstra et al. in more detail. They measured number of cycles regardless of spatial frequency. falloff and processing

concluded within

the

that low-frequency visual system.

attenuation

at the criticism of Savoy and McCann CSFs for gratings that were equated for They still observed the low-frequency in the

CSF

reflects

lateral

inhibitory

22

BANKS

AND

SALAPATEK

ages. The developmental data reported here are consistent with those two studies: 2- and 3-month-olds generally exhibited the low-frequency falloff particularly at higher luminances: l-month olds often did not (although it is possible that I-month-olds would exhibit the falloff at luminances greater than 55 cd/m”). Given the apparent relationship between the magnitude of lateral inhibition and the slope of the CSF’s low-frequency falloff (e.g., Enroth-Cugell & Robson. 1966; Van Nes & Bouman. 1965). the infant CSF results suggest that I-month-olds generally do not manifest lateral inhibitory processing but that older infants typically do. The results of Experiment 2 substantiate this viewpoint. The potential significance of postnatal development of lateral inhibition can be illustrated by considering the function of lateral inhibition in adult pattern vision. There is an astronomical amount of information in the adult’s retinal image. To simplify processing of the image. the nervous system filters and selects certain types of information. Lateral inhibition is an integral part of this filtering process. The pattern information which passes through lateral inhibitory networks appears to be, reasonably enough, that which is most useful in pattern processing: such networks tend to exaggerate or emphasize areas of transition in intensity (sharp intensity gradients such as contours) in the retinal image by attenuating or deemphasizing gradual intensity gradients (such as diffuse shadows). Since intensity transitions such as contours are the primary defining characteristics of visual patterns. the process undoubtedly facilitates adult pattern perception. Our data suggest that this contour-emphasizing process may not be present during the first weeks of life and that it may develop postnatally.

Figure 4 shows the Z-month CSF and an adult CSF obtained under the same conditions. Obviously. infants are not sensitive to nearly as wide a range of spatial frequencies as adults are (adult cutoff frequencies are typically from 30 to 50 cyideg). In addition, infants are not sensitive to nearly as broad a range of contrast values (adult peak contrast sensitivity is about 40 times greater than ?-month-olds’). In other words. the pattern information to which infants are sensitive appears to be a very small fraction of the information adults can use. Indeed the fraction seems so small as to preclude the ability to respond to all but the brightest and grossest features of visual stimulation. The limitation imposed by this small “window” does not appear quite so serious, however, when two facts are kept in mind. (I) The spatial frequencies of sine wave gratings contained within a visual object change systematically with viewing distance. As an object is brought closer to an observer. its angular size increases. Consequently. the major sine wave grating components are translated toward lower frequency values. So the infant’s “window” is

INFANTPATTERN

VISION

23

best suited to the perception of an object when it is in the immediate rather than the distant visual environment. As the “window” grows with age, the infant’s ability to perceive distant objects (or small near objects) presumably increases. (2) The contours of many common objects have contrast values greater than the threshold contrasts demonstrated by infants. For example, the contrasts of some features in the human face are typically quite high. The contrast between the first author’s forehead and hairline varies between .8 and .9 depending on lighting conditions: this contrast value is presumably high enough to be readily detected by infants (Souther & Banks, 1979). Thus. even though infant’s contrast sensitivity is low relative to adults’. it is sufficient for the detection of many typical intensity gradients in the visual environment.

To this point we have been treating the CSF as a describing function; that is, a function which elucidates some basic properties of infant pattern vision and which provides an estimate of the pattern information available. As mentioned earlier, however, the CSF can in principlr be used in a much more powerful way. It can be treated as a characterizing function: that is. a function from which the detectability of and preference for any visual pattern presented to infants can be predicted. In this section we will introduce this approach. Since it may be unfamiliar to many readers, the development will be somewhat lengthy. We have tried to develop the approach conceptually rather than mathematically to aid intelligibility. Mathematical techniques which have been developed for analyzing linear systems allow the characterizing aspect of the CSF (see Gaskill. 1978, for a readable text on these techniques). These techniques involve Fourier’s theorem which can be most easily described with reference to sounds and acoustic systems. so we will begin with a discussion of sound. Fourier’s theorem states that any sound. no matter how complex, can be reproduced by combining a set of sine waves (pure tones) of various frequencies, amplitudes, and phases. Linear systems analysis utilizes this theorem to describe the input-output relationships of linear acoustic systems. The claim is that the response of an acoustic system to any sound can be completely characterized by determining that system’s response to sine waves of various frequencies. In other words. given any sound as an input to a system, one can predict the resulting output if one knows the system’s response to the component sine waves of the sound. It must be emphasized that this is true only if the system is a linear system. the properties of which will be discussed below. The manner in which linear systems analysis describes the inputoutput relationships of (linear) acoustic systems is summarized by two equations:

24

BANKSANDSALAPATEK

@,Icf)= Q’iu-l+ % v>

121

Equation [I] describes the relation between input amplitude and output amplitude. The function A,(j) represents the amplitudes of the sine waves which compose the input sound;,f’is the frequency of the various sine wave components. Similarly, A,(f) represents the amplitudes of the sine waves which compose the output sound. AH(f) represents the amplitude of the acoustic system’s response to input sine waves with an amplitude of 1 and frequency off’. In other words, Eq. [I] states that the output amplitude of any sine wave component (with a frequency of 1000 Hz. for example) can be determined by multiplying the component’s input amplitude (A,(f)) by a weighting factor (AH(f)) which is proportional to how well the system responds to that component. The weighting factors for all frequencies are represented by A,@ which is commonly referred to as the system’s modulation transfer function. Figure 12 illustrates the application of Eq. [I] for a simple acoustic system and input stimulus. Equation [21 describes the relation between the phases of the input and output sine wave components. Q&l and cD,v) represent the phases of the output and input sine waves. respectively. (s,(j). the so-called phase transfer function. represents any phase shifts introduced by the acoustic system. Equation 121 then simply states that the phases of the output’s sine wave components can be determined by adding the phases of the input sine waves to any phase shift introduced by the system. Linear systems analysis has also been applied to optics and vision (Cornsweet, 1970; Goodman. 1968; Schade. 1956). The study of acoustic systems concerns waveforms which vary in time but the primary aim in optics and vision (particularly in the study of pattern vision) is to study the processing of patterns which vary spatially rather than temporally. The basic stimulus is now the sine wave grating. a sine wave which varies across space rather than across time. As described earlier, four terms are used to describe various sine wave gratings: spatial frequency. contrast, orientation, and phase (see Figs. 1 and 2 and associated text). Fourier‘s theorem, when applied to optics or vision, states that any two-dimensional stimulus can be exactly described by the combination of a set of sine wave gratings of various spatial frequencies, contrasts, phases. and orientations. Thus, even a complex, two-dimensional visual stimulus, such as the picture of a face. can be described exactly by the combination of a set of sine wave gratings. In Fig. 13, the Fourier or spatial frequency representation of a relatively complex pattern is shown. The pattern. a checkerboard, is shown in part A of the figure. Part B shows the coordinate system which is commonly

INFANT

FIG. depict sound

25

VISION

12. Representation of the application of linear systems analysis. The five graphs in sequence how linear systems analysis can be used to predict how a particular input (A) will be passed by an acoustic system (C) to yield an output sound (E). (A) The

waveform

of the

input

sound.

spectrum of the input sound. (pure tones) which comprise function. The amplitude this function represents The amplitude various input

spectrum sine waves

was obtained

by using

waves by the response sound. Note how the particular, the ripples system

PATTERN

does

not

respond

Intensity

is plotted

This spectrum the input sound.

of response to sine how well the system

waves passes

of the output sound. once they have passed Eq.

[ 11; that

as a function

with sine

is, by multiplying

to higher

(B)

The

the

amplitude sine

waves transfer

an amplitude of I is shown. waves of different frequencies.

The spectrum shows through the acoustic

amplitude of the acoustic system. waveform has changed in passing in the input waveform have been well

of time.

shows the amplitudes of the various (C) The acoustic system’s modulation

amplitude (E) The through smoothed

the amplitudes system. This of each

of the

waveform the acoustic because

Thus, (D) of the spectrum

input

sine

of the output system. In the acoustic

frequencies.

used to display the sine wave grating components of a two-dimensional pattern (see figure caption for more information). Part C is the actual Fourier or spatial frequency representation of the checkerboard, which is obtained from Fourier transformation of the pattern (see Gaskill, 1978, pp. 1 I l-l 13). This spatial frequency representation is called an amplitude spectrum. It contains all the information needed to specify the spatial frequency, contrast, and orientation of the checkerboard’s constituent sine wave gratings. Each spot represents an individual sine wave compo-

36

BANKS

AND

L-imL-

SALAPATEK

L SPATIAL

FREQUENCY

. .._ \ . . . ._ . . . .,‘. . . . ,. .*~,F ..* x~L. . . . . 4,: . .i-. ,. , . .. . . . .1.

,L_L ,-l-lSPATIAL FREQUENCY FIG.

13.

A checkerboard

and

its

spatial

(CI/OEG)

frequency

representation.

checkerboard often used

composed of equal width light and dark to display the sine wave grating components

coordinate the origin:

system. the spatial frequency of a component is represented that is. spatial frequency is proportional to the length of the

to the component. The orientation of the component is vector and the positive horizontal axis. (Cl The spatial spectrum) of the checkerboard. Each spot represents Contrast or amplitude is represented by the area of representation (amplitude spectrum) of Contrast or amplitude of each component diagonal components plotted are indicated

(A)

elements. (B) The of a two-dimensional

represented frequency an individual the spot.

An

coordinate pattern.

II

x

12

system In this

by its distance from vector from the origin

by the angle representation sine wave (D) The spatial

between the (amplitude component. frequency

the diagonal components of the checkerboard. is plotted as a function of spatial frequency. by the broken line in part (Ct

The

nent. Spatial frequency is represented by the spot’s distance from the origin of the plot; the greater the distance. the higher the frequency. Contrast or amplitude is represented by the area of the spot and orientation by the angle between the horizontal axis and a line from the origin to the spot. The point is that this amplitude spectrum provides an exact description of the checkerboard in terms of its constituent spatial frequency components (technically the phase spectrum is also needed. but

INFANT

PATTERN

VISION

27

for simplicity we will consider only the amplitude spectrum henceforth). Part D of the figure shows another way of representing an amplitude spectrum which will be used later in the paper. The graph represents the diagonally oriented frequency components of the checkerboard. (These components are surrounded by a broken line in part C of the figure.) Spatial frequency is represented by the horizontal axis and amplitude or contrast by the vertical axis. Keeping things simple, we will first consider how linear systems analysis can be used to characterize the image-forming capabilities of an optical system such as a camera lens before considering how it can be applied to vision. Linear systems analysis allows one to predict the image (output) produced by a lens when presented any two-dimensional pattern (input) if one knows the lens’ modulation transfer function. The manner in which such input-output relations are calculated is summarized by twodimensional equations similar to the one-dimensional Eqs. [I] and [2].

Equation [3] relates the amplitudes of the input and output spatial frequency (sine wave gratings) components. A,cf,g) is the optical system’s two-dimensional modulation transfer function; it describes how well sine wave gratings of different spatial frequencies are transferred into the image formed by the lens. The modulation transfer function of most camera lenses indicates that low and medium frequency gratings suffer no significant loss of contrast and, therefore, are well represented in the image or output. Higher-frequency gratings suffer progressively greater contrast reduction. Equation [43 relates the phases of the input and output spatial frequency components. Fortunately. it is generally assumed that @&,g), the lens’ phase transfer function, is zero for all spatial frequencies so that Q”cf;g) is equal to @,cf,g). (This assumption simply means that the lens does not laterally displace some sine wave gratings and not others.) We will follow this assumption and henceforth disregard optical (and visual) system’s phase transfer functions (see Cornsweet. 1970. pp. 323-324). The linear systems approach outlined by Eq. [31 has proven to be very useful in assessing the image-forming qualities of lenses. Linear systems analysis has also been used to study visual systems (Cornsweet, 1970; Davidson. 1966; Ratliff, 1965). If a visual system is linear. linear systems analysis claims that the perceptual response to any visual pattern can be predicted if one knows the system’s sensitivity to sine waves of various spatial frequencies. This sensitivity function is actually the CSF (contrast

28

BANKSANDSALAPATEK

sensitivity function). Thus, in principle. the CSF provides a complete characterization of how system processes two-dimensional patterns if the system is a linear system. For a system to be classified a linear system, four assumptions must be linearity. and homogeneity. The satisfied: isotropy. state-invariance. adult visual system does not satisfy any of these exactly; thus, the CSF cannot completely characterize how the adult visual system represents two-dimensional stimuli. There is no compelling reason to expect that the infant visual system exactly satisfies these four assumptions either. However, it is possible to minimize the inaccuracies that arise from violating the assumptions by measuring CSFs under various conditions and by restricting the application of linear systems analysis to those situations in which the violations are small. We will now discuss each assumption in order to characterize those situations. Isotropic optical or visual systems are those which respond similarly to input stimuli of any orientation. The adult visual system is not isotropic since visual acuity is higher for vertically and horizontally oriented targets (gratings or single bars) than it is for obliquely oriented ones (Appelle, 1972). Interestingly, the young infant’s visual system may be isotropic. Teller et (11. (1974) and Gwiazda rt ~1. (1978) found that 2- to 5-month acuity is similar for vertical. horizontal, and oblique gratings. Leehey et al. (1975) have reported evidence to the contrary, their infants, aged 2 to 12 months, exhibited higher acuity for vertical and horizontal gratings. The problem of anisotropy can be virtually eliminated by measuring the CSF for stimuli presented in various orientations and utilizing this orientation-specific information when applying linear systems analysis. The assumption of state-invariance requires that the response function of the visual system in question not vary with adaptation level. The adult visual system does not satisfy this requirement either. Very dim. scotopic, stimuli are processed differently than bright, photopic ones. Indeed, the CSF of the adult visual system varies significantly with space-average luminance or adaptation level (Van Nes & Bouman. 1965: Schade, 1956). This problem can be avoided by measuring CSFs at a fixed level of average luminance and applying linear systems analysis to only those stimuli of similar average luminance. Thus, a number of CSFs at various average luminances should be obtained to allow analysis of stimuli of different luminances. The property of linearity concerns the relationship between input and output amplitudes. If linearity holds, the output or response of the system under study is a linear function of the input amplitude; that is to say, if the input amplitude is doubled, the output amplitude also doubles. The adult visual system does not appear to be linear; response is often better approximated by a logarithmic than by a linear function of input ampli-

INFANTPATTERN

VISION

29

tude. Violation of the assumption of linearity is greatest when large intensity gradients are present in the stimulus (Cornsweet, 1970). Consequently, the normal practice among visual scientists is to use lowcontrast stimuli so that linearity is not significantly violated. Thus, systems analysis seems best suited to the study of near-threshold patterns; in other words, it is best suited to the study of pattern detection. The property of homogeneity requires that the response of the system is the same regardless of the spatial location of the stimulus. In the case of the visual system, it requires that the CSF be the same regardless of the retinal region stimulated. This assumption is, of course, also violated by the adult visual system since the response function of the retina varies dramatically from the fovea to the periphery (Hilz & Cavonius, 1974). The best approach is to use a relatively large stimulus and assume that the central retina, being more sensitive than the periphery, determines the contrast threshold obtained (Wilson & Geise, 1977). Then one assumes that the CSF for large-field gratings represents the sensitivity of central vision. The point of the above discussion is that one may very well be able to use the CSF to make predictions about the infants’ responses to a large variety of visual patterns under somewhat restricted conditions. The primary restrictions are: (I) use of visual stimuli whose contrast is not significantly above threshold; (2) use of stimuli whose average luminance does not vary across time; and (3) caution concerning which segment of the visual field is actually being studied. Given that these restrictions are acknowledged, the CSF and linear systems analysis may be quite useful in the attempt to characterize infant pattern vision. In the next two sections. we will present three examples to illustrate this. Application of Linear Systems Analysis to Existing Data on Infant Pattern Detection The CSF can be used to approximately define a very important aspect of an infant’s pattern vision: the boundary between what is visible (detectable) and what is invisible (undetectable). It is reasonable to assume that the CSF is a good estimate of this boundary since at or near threshold the property of linearity is least violated if at all. This means that the CSF and linear systems analysis may be very valuable to the characterization of pattern detection in young infants; one can use the approach to calculate whether a particular pattern is sub- or suprathreshold to infants of various ages. In this section, we present two examples in which the results of infant pattern detection studies were predicted by using the infant CSFs and systems analysis. The successful predictions illustrate the potential value of linear system analysis to the study of infant pattern detection.

30

BANKS

1

a

b

c

d

AND

SALAPATEK

2

INFANT

PATTERN

31

VISION

Dayton et al. (1964) measured newborn visual acuity using drifting square wave gratings to elicit optokinetic nystagmus (OKN). They reported acuity thresholds between 1.5 and 4 cyideg (20 and 7.5 min of arc), values which are notably higher than those reported by other investigators (Gorman et al., 19.57, 1959; Miranda, 1970). The CSF and linear systems approach provides insight into why the acuity values they report are so much higher than the estimates of other investigators. To demonstrate this, we performed the analysis depicted in Fig. 14. The general outline of linear systems analysis as used in such analyses is the following. A particular pattern (part la. for example) is Fourier analyzed to obtain its amplitude spectrum (part lb). The amplitude spectrum is then filtered by (multiplied by) a function similar to the infant CSF to obtain the filtered or output amplitude spectrum (part Ic). (The function we used to represent the infant CSF passed spatial frequencies less than I cyideg and filtered out frequencies greater than 1 cy/deg.) Finally, the output amplitude spectrum undergoes Fourier synthesis to obtain the filtered or output pattern (part Id). One way to think of the whole process is that the pattern is being filtered in the spatial frequency domain to ascertain what information from it is able to pass through the “window’* of the CSF. Sections la and 2a of the figure are photographs of regularly spaced and irregularly spaced gratings, respectively. The regularly spaced grating was produced mechanically so the individual bars are nearly identical in width. The irregularly spaced grating is a near replica of the pattern used by Dayton et al. Their pattern and our replica were produced by handpainting black bars on a white background. Sections lb and 2b show the input amplitude spectra of the regular and irregular gratings, respectively. The regularly spaced grating has a spatial frequency component of high

FIG. ings.

14. Spatial Two patterns

composed posed of irregular gratings

frequency are shown

of regularly irregularly gratings because

spaced, spaced

are identical. it was most

representations in the upper light stripes The similar

of regular and irregular square wave gratportion of the figure: a square wave grating

and dark stripes ( la) and (2a). The average stripe irregular grating was chosen to the grating used by Dayton

a square widths

wave grating of the regular

from a series et nl. (1964)

I). The spatial frequency representations (amplitude spectra) for in parts lb and 2b. (The amplitude spectrum of the regular grating at 1.5. 3, 4.5. 6, and 7.5 cyideg whereas a true square wave (see components at 1.5, 4.5, and 7.5 cyideg. This discrepancy is photographic reproduction of the grating (la) which caused the wider than the light stripes.) Parts Ic and 2c depict the spatial

comand

of hand-drawn (see their Fig.

the two gratings are shown has sine wave components Fig. 15) would only have due to “bleeding” in the dark stripes to be slightly frequency representations

(output amplitude spectra) of the two gratings once frequencies above and including 1 cyideg have been filtered. Parts Id and 2d show the patterns which are obtained from reconstruction (Fourier synthesis) of the output spectra. The photographs shown in rows b. c, and d were produced by a coherent, frequency-domain filtering system operated by the Department of Electrical system

see

Engineering Becker

and

at the University Knopp (197X).

of Texas.

For

a description

of this

optical

filtering

32

BANKS

AND

SALAPATEK

amplitude at 1.5 cy/deg.7 This component is the fundamental of the grating. One can see that the amplitude spectra of the gratings differ, the regular grating contains frequency components tightly grouped around 1.5. 3, 4.5 and 6 cy/deg whereas the irregular grating has a greater dispersion of frequency components. Sections Ic and 2c depict the output amplitude spectra associated with the two patterns once all of the sine wave components with spatial frequencies of 1 cyideg or greater have been filtered out. These output spectra are the product of the input spectra ( I b and 2b) and a CSF with a cutoff frequency of the I cyideg (see Eq. 131). Note that the output spectrum associated with the regular grating contains only very low amplitude frequency components (except for a prominent component with a spatial frequency of zero which corresponds to the average luminance of the original pattern). In other words. once the original stimulus is filtered through the CSF, very little information remains. This is because the visual system, represented by the CSF we used, is not sensitive to spatial frequencies as high as those in the regularly spaced grating. On the other hand. the output spectrum associated with the irregularly spaced grating indicates that considerable information from this grating is able to pass through the CSF; it contains frequency components within the sensitivity range of the idealized CSF. Sections I and 2d illuminate the significance of this most clearly. These photographs are the result of constructing the original gratings from the output spectra of sections I and 2c. In other words, they are reproductions of the original stimuli once they have been filtered by the CSF. Obviously, the pattern within the regular grating has been attentuated to near threshold levels. However, the pattern within the irregular grating, although significantly attenuated, is still clearly visible. This illustrates then how Dayton et al. could have overestimated newborn acuity. If we assume that newborn acuity is about 1 cy/deg (20/600) as others have (Gorman et al., 1957. 1959; Miranda. 1970) one can see that newborns could still respond to an irregular grating with a fundamental frequency of 1.5 cyideg or greater. We have performed the same analysis for an irregularly spaced grating with a fundamental of 4 cy/deg and found that some pattern information lower than I cy/deg was still present. The CSF and linear systems analysis in this case provide a means to relate seemingly contradictory findings, a means based on a logical and effective manner of describing visual stimuli. ’ At a later point we state that the amplitude spectrum of a regularly spaced. square wave grating consists of the fundamental plus all of the odd harmonics; that is. it consists of the fundamental (f), the third harmonic (30. the fifth harmonic (Sf). and so forth. Note that the amplitude spectrum in part ib of Fig. 14 contains odd and even harmonics even though the grating (la) appears stripes are not exactly harmonics.

to be a square wave grating. the same width; this leads

In actuality the to the introduction

grating’s of even

light and dark as well as odd

INFANT

PATTERN

VISION

33

There is another seemingly puzzling finding in the infant perceptual literature which can be understood from the CSF point of view. Fantz et al. (1975) used two types of stimuli to estimate infant acuity. Examples of these stimuli are shown in Fig. 15. One stimulus was a regular square wave grating composed of equally spaced light and dark stripes. The other was a grating composed of six dark stripes separated by wider light stripes. This type of pattern is commonly termed a rectangular wave grating. Fantz et al. performed two detection experiments, one with each kind of stimulus. In both experiments, the gratings were paired with a SQUARE-

WAVE

RECTANGULAR-WAVE

nII

GRATINGS

GRATINGS

FIG. 15. Examples of square wave and rectangular wave gratings used by Fantz et al. (1975). Two square wave gratings, one with a stripe width of I5 min of arc and the other with a stripe width of 7.5 min of arc, are shown to the upper left. Their spatial frequency representations (amplitude spectra) are shown to the upper right. Two rectangular wave gratings, one with a stripe width of 7.5 min of arc and the other with a stripe width of 3.7 min of arc are shown to the lower left. Their spatial frequency representations (amplitude spectra) are shown to the lower right.

34

BANKS

AND

SALAPATEK

constant luminance (unpatterned) stimulus. Visual acuity was estimated for I-, 2-. and 3 &month-olds by determining the finest gratings for which significant fixation preferences could be obtained. However. fineness was varied in different ways for the two gratings. The square wave grating underwent identical changes in the width of both the light and dark stripes. Thus. both the number of stripes and the stripe widths varied with changes in fineness. On the other hand, the fineness of the rectangular wave grating was increased by decreasing the width of the dark stripes without changing the number of stripes in the pattern (the light stripes were therefore widened somewhat). Figure IS shows two examples 01 each grating. The upper examples are two square wave gratings with different stripe widths while the lower examples are two rectangular wave gratings with different stripe widths. The two types of gratings yielded apparently different acuity estimates. Fantz rt ~1. state that “finer lines were detected against a plain ground (the rectangular wave grating) than were resolved in patterns of striations (the square wave grating)” (p. 266). These results are shown in Fig. 16: the open squares representing the finest stripes detectable with the square wave and the open circles the finest stripes detectable with the rectangular wave. On the surface it seems a peculiar result that narrower stripe widths were detectable when they were widely spaced than when they were more

I I

2L

I 3

2

I 4

AGE ,MONTHS~ FIG.

16.

Comparison

of

observed

and

predicted

detectability

of

square

wave

and

rec-

tangular wave gratings. The stripe width (in minutes of arc) of individual dark stripes is plotted as a function of age. The open symbols represent the minimum detectable stripe widths reported by Fantz et ul. (197% for square wave gratings (0) and rectangular wave gratings (0). The filled symbols represent our predictions of minimum detectable stripe width for square wave (U) and rectangular wave gratings (0). See text for an explanation of the predictions.

INFANT

PATTERN

VISION

35

closely spaced. Applying the CSFs of Fig. 8 and a form of linear systems analysis to these data we have demonstrated that the results from the two experiments are in fact congruent. As in the first example, the stimuli were Fourier analyzed to determine their amplitude spectra; the spectra of the square and rectangular wave gratings are depicted in the right half of Fig. 15. The square wave possesses several spatial frequency components but the component of greatest amplitude has a spatial frequency of l/X where X is the distance between the midpoints of adjacent dark stripes. This component is the fundamental of the square wave. Note that a decrease in stripe width from 15 mm to 7.5 min leads to a reduction in X and, consequently, to an increase in the frequency of the fundamental, l/X. The fundamental of the rectangular wave grating is also at I/X. However. since the distance between adjacent dark stripes is constant when stripe width is varied. the fundamental of the grating is constant at l/X cyideg for all stripe widths. Decreases in stripe width change the amplitudes of the various frequency components but not their spatial frequency. Using these amplitude spectra, we determined the results that would be expected from the CSFs of Fig. 8. We assumed that infants’ detection of the gratings was determined by their ability to detect the gratings’ fundamental frequency component since it is the component of greatest amplitude. For the square wave grating, the contrast of the fundamental was approximately 417r (0.5) or 0.635. We then calculated the spatial frequency that would be just at threshold when presented at a contrast of .635. The l-month-old CSF indicates that this occurs at about 1.9 cyideg for I-month-olds. The two-month CSF indicates that it occurs at about 2.35 cyideg and the 3-month CSF. at about 3.3 cyideg. We converted these values into stripe widths and the resulting predictions are shown as solid squares in Fig. 16. The fit between these predictions and the observations of Fantz et (11. is quite good, particularly at 2 and 3+ months. The analysis was also applied to the rectangular wave. In the Fantz et ul. study the fundamental of this pattern was 0.35 cy/deg regardless of stripe width. Again we assumed the infants’ detection of the pattern was determined by their sensitivity to this fundamental. We used each age group’s CSF to calculate the minimal stripe widths that would yield threshold amplitude at 0.35 cyideg. These predictions are shown as solid circles in Fig. 16. The fit between predictions based on infant CSFs and the observed data of Fantz et al. is again quite reasonable. More importantly. the approach predicts that finer stripe widths should be detected with rectangular wave gratings than with square waves. Thus, it is not necessary to postulate differences in the manner of processing the two patterns to explain the original finding of different minimal stripe widths because this divergence in results can be predicted quite accurately from a single function: the CSF.

36

BANKS AND SALAPATEK

Application oj’ Linear Systems Analysis to Existing Data on Injbnt Pattern Preferences Both of the examples in the last section were drawn from pattern detection experiments which. as mentioned earlier. should be quite amenable to the linear systems analysis approach. However, for the approach to be of more general value to the study of infant pattern vision, its descriptive and predictive validity should extend to more complex. suprathreshold patterns. There is, of course, a vast literature concerning infants’ discrimination of and preference for suprathreshold patterns. One would not anticipate that systems analysis could yield highly accurate predictions for most of these experiments because the use of suprathreshold patterns may lead to violations of the linearity assumption. Nonetheless, one might be able to make predictions of reasonable accuracy in such situations and thereby account for a significant proportion of the findings. With this thought in mind, we applied the CSF and a form of linear systems analysis to a series of infant preference studies which employed presumably suprathreshold patterns. Several investigators have found that infants preferentially fixate some checkerboards over others depending on the size of the checks (Brennan et al., 1966; Cohen, 1972; Greenberg & O’Donnell, 1972; Karmel, 1969a, 1969b; Maisel & Karmei, 1973). Interestingly, the maximally preferred check size appears to decrease with age. Karmel and Maisel(l975) have nicely summarized these findings. Figure 17 is based on their Fig. 2.3; it depicts the relationship between maximally preferred check size and age as reported in a number of studies. Kelly (1976) has derived the amplitude spectra of checkerboards ot various check sizes (see Fig. 13). These spectra reveal that checkerboards contain a number of spatial frequency components at a number of orientations. The component of highest amplitude, the fundamental, has a spatial frequency of @D, where D is check size (in degrees). Using the checkerboard amplitude spectra, we applied a form of systems analysis and our infant CSF data to predict maximally preferred check size for different age groups. We assumed again that visual preferences are governed by processing of the fundamental frequency. Given this assumption, the maximally preferred checkerboard for a given age group should have its fundamental frequency component at the peak of the age group’s CSF. We fit each of the CSFs (from Fig. 8) with smooth functions and found that the peaks were at 0.20 cy/deg for I-month-olds, 0.38 for 2-montholds, and 0.57 for 3-month-olds. Therefore, we predict that the three age groups should maximally prefer checkerboards with fundamentals of 0.20. 0.38, and 0.57 cyideg. These predictions, shown as open squares in Fig. 17, clearly match the observed preferences quite well.

INFANT

lot

'

PATTERN

/ 2

/ 3 AGE

37

VISION

I 4

I 5

,

( MONTHS1

FIG. 17. Comparison of observed and predicted checkerboard preferences. Maximally preferred check size is plotted as a function of age. The filled symbols represent the results of a number of different experiments as analyzed by Karmel and Maisel(1975) (see their Fig. 2.3). The open symbols represent our predictions of maximally preferred check size at I, 2, and 3 months. See text for an explanation of the predictions. Most of the filled symbols were obtained from experiments in which only checkerboards were presented (Brennan et al.. 1966; Cohen, 1972; Hershenson, 1964). Some of the data points, however. were obtained from experiments which used checkerboards and other related patterns. In two cases (Greenberg & O’Donnell, 1972; Maisel & Karmel, l973), we were unable to isolate the checkerboard data and, consequently, the data points represent combined preferences for checkerboards and related patterns.

This result further substantiates the potential utility of the CSF approach for describing and understanding infant pattern preferences. It is particularly important for two reasons. First, it shows that the predictive power of the CSF extends beyond the type of patterns used to measure it; in other words, the CSF determined with linear sine wave gratings can be used to make relatively accurate predictions concerning nongrating patterns. Second, the result suggests that the CSF and linear systems approach may have predictive power for the suprathreshold patterns used in most infant preference studies. Summary

of the CSF Approach

to Infant Pattern

Vision

The CSF and forms of linear systems analysis have been very useful in the study of adult pattern vision. We have proposed that the CSF approach could also contribute significantly to our understanding of visual development in two genera1 ways; its use as describing pattern vision capabilities and its use as characterizing pattern capabilities.

38

BANKSANDSALAPATEK

When regarded as a describing function. the CSF illuminates some very basic aspects of pattern vision: visual acuity. sensitivity to contrast. and low-frequency attenuation. We have discussed the importance of each of these toward an understanding of the way infants see. The characterization aspect of the CSF. when developed more fully. could potentially yield a fundamental synthesis of much of the research on pattern vision in young infants. The CSF and linear systems analysis can be legitimately applied to studies in which pattern detection thresholds have been measured, i.e., in studies where the assumptions of systems analysis are least violated. We have shown that the approach adequately predicts infant detection thresholds for different types of gratings with two examples of systems analysis applications (Dayton rt rrl., 1964; Fantz et al., 1975). In future work, systems analysis should be applied to other studies of pattern detection (e.g.. detection of single bars. curvilinear targets. etc.). Most of the literature on infant pattern vision, however. concerns pattern discrimination which in habituation or preference experiments involve suprathreshold patterns. We do not believe that the CSF approach can at present provide any general account of the pattern ~ii.~cri~~ination capabilities of young infants. Nonetheless. it may be very useful in interpreting three sorts of pattern discrimination findings. One could determine. to a first approximation. whether the two patterns in a discrimination experiment were in fact suprathreshold. In one case. if neither pattern were detectable, no discrimination would be predicted. In a second case. if one pattern were detectable and the other were not, discrimination would be predicted. The CSF may aid the understanding of infant pattern discrimination in yet a third situation. Numerous physically dissimilar suprathreshold patterns. when filtered by the infant CSF. may yield identical output amplitude spectra: that is, the pattern information which differentiates such patterns might not be detectable. In such cases. no discrimination would be predicted. Thus, one could use infant CSFs to determine classes of perceptually equivalent. yet physically dissimilar patterns. Campbell and Robson (1968) have used similar logic to accurately predict conditions under which adults cannot discriminate a square wave grating from a sine wave grating. Despite these important implications, the CSF approach is not well suited to a general description of the discrimination capabilities of young infants. We believe, nonetheless, that the approach should be an integral component of a general theory of infant pattern discrimination. The CSF approach may be more valuable to the study of infant pattern preferences per se. Our accurate predictions of age changes in maximally preferred check size support this claim. A simple model of infant preference can be proposed at this point: one must keep in mind, however. that the model cannot be entirely accurate if the assumptions of linear systems

INFANT

PATTERN

VISION

39

analysis are seriously violated. The mode1 assumes that infant preferential looking is governed by the pattern information available to “decision centers” in the central nervous system. There are two important facets of this genera1 assumption. First, the pattern information upon which the decision centers operate is presumably only a small fraction of the information impinging on the infant’s eye: considerable information is lost in optical and neural processing by the ocular media and visual pathways. We assume that the CSF approach offers a way to determine what pattern information is in fact conveyed to the decision centers. Second, we assume that relatively simple decision rules are applied to this remaining information. Three rules seem most reasonable a priori. ( I) Young infants will preferentially fixate the patterned stimulus which possesses the spatial frequency component of highest amplitude in the output amplitude spectrum. This is the rule which was used in the third example because it is the simplest of the three to explicate. (2) Young infants will preferentially fixate the stimulus which possesses the greatest total energy in the output amplitude spectrum. In other words, decisions will be based on the magnitude of the integral of (area under) the output spectrum ( !“,A o(f’)dj”f). (3) Young infants will preferentially fixate the stimulus which, once filtered, possesses the greatest amount of contour density. Contour density in this case must be defined in a nonstandard way. It refers to the total length of contour per unit of area multiplied by the contrasts of each of the contours. (One might consider the following equation to implement this:

where g(x,.v) is the inverse Fourier transform of the output amplitude and phase spectra and I: is the average luminance of the pattern.) Unfortunately, the patterns we have investigated to date do not allow us to differentiate these three decision rules. In future research we hope to analyze preferences among particular patterns which will allow such differentiation. Regardless of the decision rule chosen, the model predicts developmental changes in pattern preferences due to changes in the pattern information conveyed to decision centers. This basic model is not the first to propose that very simple mechanisms govern early pattern preferences. Two classes of models have appeared in the developmental literature: (I) models which propose a simple neural substrate for preferences and (2) models which propose specific stimulus dimensions as predictors of early preference. The neural substrate models have proposed that young infants‘ visual behavior is governed by the responses of neurons in the central visual system (Fantz et al., 1975; Haith. 1978: Karmel & Maisel. 1975). Thus.

40

BANKS

AND

SALAPATEK

the infant’s “window” is associated with the receptive-field properties of cortical (or subcortical) neurons. The model further proposes that these receptive fields might become more highly tuned with age and. consequently, older infants’ preferences tend toward finer patterns. Unfortunately, these models in present form cannot be verified experimentally since the “window” they postulate cannot as yet be independently measured in human infants; physiological data from other species are required to characterize developmental change in the size and shape of cortical (or subcortical) receptive fields. Hopefully, when data from closely related species such as nonhuman primates become available. testing of such models will become more feasible. In contrast, the “window” we have proposed is represented by the CSF: thus, its size and shape can be determined from behavioral rather than physiological observations. The CSF “window” may be very closely related to the visual cortex “window” proposed by Haith (1978) and Karmel and Maisel (1975); nevertheless, our approach has distinct advantages since one can empirically determine the CSF for infants at various ages and use such data to predict the detectability of and preference for a wide variety of patterns. In short, the present approach is capable of more quantitative hypotheses and, therefore, is more amenable to empirical test. The second class of general models that has been proposed claims that particular stimulus dimensions are fundamental to infant pattern preferences. Several unidimensional models have been proposed (e.g., Hershenson et al., 1965; Kagan, 1970). We will only discuss one of these models-Karmel’s (1969a)--since it is the most specific of this group. Karmel (1969a) proposed that contour density, which is defined as the total length of contour per unit of area in a visual pattern, is the primary determinant of infant pattern preferences. He points out that other stimulus dimensions such as motion, depth, and average intensity may also be determinants of preference, but these are not considered dimensions of pattern given our working definition, so we will restrict our discussion to dimensions relevant to patterns per se. Karmel(1969a) showed that visual preferences for checkerboards are described by an inverted U-shaped function of contour density; that is, checkerboards of intermediate check size are most preferred. Moreover, the U-shaped function varies from one age to another, maximally preferred contour density increasing with age. Of most interest is the relationship between contour density and preferences for other sorts of patterns. Karmel (1969a) demonstrated that one can use his U-shaped functions relating preference to contour density to predict relative preferences for random checkerboards as well. In a later publication, Karmel and Maisel (1975) showed that the same U-shaped functions predict relative preferences for fields of dots of various size and square wave gratings. The contour density model of infant pattern preferences bears some

INFANT

PATTERN

VISION

41

similarity to the model presented here. (Karmel and Maisel have anticipated this possibility [Karmel & Maisel, 1975, p. 841.) It is easy to show, however, that the present CSF model is likely to have greater predictive validity than Karmel’s contour density model. There are a large number of different patterns which have the same contour density. The rectangular wave gratings used by Fantz et al. (1975) (and in our second example of applying the CSF approach) are good examples of such stimuli. As one varies the stripe width of rectangular wave gratings, their contour density does not vary significantly. Thus, the contour density model would predict that infants’ responses to these patterns would be relatively independent of stripe width. Of course, this prediction is not borne out; fixation of the gratings declines significantly for smaller stripe widths. The CSF model was able to predict such changes. The general problem with Karmel’s model appears to be that contour density does not incorporate enough of the pattern dimensions that are important to early visual behavior. We conclude that our CSF model, although similar in some respects to Karmel’s contour density model, should be more capable of generally characterizing infant pattern preferences since it is based on a more effective means of describing patterns and visual systems’ processing capabilities. Conclusion

We have argued that much of early preference behavior can be characterized by the CSF approach. This assumes that early visual preferences are determined primarily by the pattern information available to decision centers in the central nervous system and that these decision centers operate upon this information in a relatively automatic manner. Our approach may allow one to estimate what pattern information is actually supplied to such centers and how this information changes with age. We proposed a simple model of infant pattern detection and pattern preferences based on this hypothesis. This model is only a subunit of the type of model one might propose to account for visual selectivity in older infants, children, or adults. An account of selectivity in older subjects would assuredly have to consider the contributions of selective attention and long-term memory since certain stimulus configurations probably become very salient if subjects learn from previous experience that the configurations possess particular meaning. Thus, processing stages beyond the peripheral stages we believe the CSF approach captures would have to be incorporated. This is not to say that our approach would be irrelevant to such a model; the approach is undoubtedly useful in describing the filtering of pattern information supplied to higher stages of processing in older children and adults as well as young infants. If the CSF approach proves to be successful in describing early preference behavior, it will be of particular interest in the future to determine the ages for which the CSF

42

BANKS

AND

SALAPATEK

approach does not adequately account for pattern preferences since the failure to account for preference behavior will imply that processes other than simple optical and neural filtering by the “window” of the visual system have become significant to the infant’s perceptual world REFERENCES Allen.

J. \‘i.sunl doctoral

dc\vlopmrnr

crcuity dissertation.

Appelle. S. Perception effect in man and Atkinson.

0..

(London).

Atkinson. sured 1979.

18, J..

II,> to 6 month.\

& Braddick.

F. Acuity

of stimulus

orientation:

Banks. Barlow,

0.. & evoked

Braddick. and static

French, potential.

0.. and Moar. patterns. Vision

1980,

S., &

C’ision M. human

S.. & infants.

and

contrast

sensitivity

J. Contrast srnsltivity of the in~~c,sri,~~~ri~r Ophthulmolr~g~ K.

Contrast

Resrorch.

M.

F.,

sensitivity

1977.

&

E. The

Knopp,

and

of infant

vision.

human

neonate

trnd Visrtul

of

the

human

17, 1045-1047.

function

contrast

0~~1ztlrn/nlolo,~?

J. Processing

cd

influence

The

sensitivity

meaSci~~ncr.

Infant

for

(a)

sensitivity ovet- the tirst 3 17. 1037-1044. (h) early infancy. Child De-

of

visual

P.s~c.hr~pt~~.rit~,\, of the albedo

Jotyntrl

and

of the

sensitivity

cd

R.. & Kuffler, S. W. Change during dark adaptation. Journal

Perc~eption

G. W.

P. Acuity

Inwsriguriw

human infant. Brifish Brennan, W. M., Ames, patterns of different Bronson.

P. Contrast 16, 867-869.

Salapatek.

H. B.. Fitzhugh. of the cat’s retina

domains. Berlyne. D.

oblique

51, 646-666.

Salapatek. Re.srarch. 1976.

Becker,

The

ZIO-213.

l,e/opnwnt, M.

Unpublished

1977. 78, 266-178.

Atkinson. J.. Braddick. 0.. & Moar, K. Development ofcontrast months of life in the human infant. I’i.Gon Rrseurch. 1977. Banks. M. S. The development of visual accommodation during Banks.

ctf crgr.

197X.

1974. 247, 403-404

J.. Braddick. by the visual

Atkinson, moving

it(firnts

of Washington.

and discrimination as a fun&on animals. I’syc-ho/o~cc~u/ Bnllrtin.

.I.. Braddick.

Nature

in human

University

Vi.otal

in

of organization of’ Physiology, illusions

1978. 23, complexity

of visual

in the

Child

3-month-old

17, 361-365.

and spatial

frequency

52 I-576. of stimuli

capacity.

lY7X.

system.

in the receptive field\ 1957. 137, 327-337.

on visual

of’ P.sdw/og~.

growth

visual

I-. 2.. and

Scicv~~c~.

19%. 56, 3 15-3 E. W.. & Moore, R. W. Age differences complexity. Sc~ienw. 1966. 151, 354-356.

postnatal

infant

fixation

IX. in infants‘

in the

attention

Dc~~~clopment.

to

lY74. 45,

X73-890. Brown.

J. I... & Mueller. C. B. Brightness discrimination and brightness Graham (Ed.). Vision und risrtul percuptiorl. New York: Wiley. Campbell. F. W.. Carpenter. R. H. S.. & Levinson. J. Z. Visibility compared with that of sinusoidal gratings. Jrntmul ofPhysiology.

Campbell. F. W.. & Green, D. G. Optical and Jurtrnnl uf‘ Physiology, 1965. 181, 576-593. Campbell, F. W.. & Robson. J. Cr. Application gratings. Jolrrnul of‘ Physidogy. 1968. 197, Cohen. L. Attention-getting and attention-holding Child Dt~vdopmrnt. 1972. 43, 869-879.

retinal

factors

of Fourier 5S l-566. processes

contrast. 1965. of aperiodic 1969. 204,

affecting

visual

analysis

to

in

visual

infant

the

In C. H. pattel-ns 183-798. resolution.

vlsibdity

of

preferences.

Cornsweet. T. N. Visd Percrptiorr. New York: Academic Press. 1970. Davidson. M. L. A prrtrrrhution unulysis of‘sputiul brightness intrrwtions in,flushrd ~~isuul fields. Unpublished doctoral dissertation, University of California at Berkeley. 1966. Dayton, G. 0.. Jones. M. M.. Aiu. P.. Rawson. R. A.. Steele. B.. & Roar. M. Developmental study of coordinated eye movements in the human infant I: Visual acuity in the newborn human: A study hased on induced optokinetic nystagmus recorded by electrooculography. Archives ofOphthulmo10~~~. 1964. 71, 865-870.

INFANT

PATTERN

VISION

43

Dobson. V.. & Teller. D. Y. Visual acuity in human infants: A review and comparison of behavioral and electrophysiological studies. Vision Research. 1978. 18, l469- 1483. Doris, J., Casper, M.. & Poresky, R. Differential brightness thresholds in infancy. Journnl

c~f‘Experimenta1 Child Psychology, 1%7. 5, 521-535. Doris,

J.. &Cooper.

L. Brightness

discrimination

in infancy.

Jonrnul ofExperimentu/

Child

P.s~cholog~. 1%6. 3, 31-39. Enroth-Cugell, C.. & Robson. J. G. The contrast sensitivity of retinal ganglion cells of the cat. Journal of Physiology. 1966, 187, 517-552. Estevez. 0.. & Cavonius, C. R. Low-frequency attenuation in the detection of gratings: Sorting out the artifacts. Vision Rrsrurch, 1976. 16, 497-500. Fantz. R. L. Pattern vision in young infants. Psyho/ogica/ Recurd. 19SX. 8, 43-47. Fantz. R. L. Visual perception from birth as shown by pattern selectivity. In H. E. Whipple (Ed.). New issues in infant development. Anna/.s of NW York Academy of’ Science. 1%5. 118, 793-814. Fantz. R. L., Fagan. J. F., III. & Miranda. S. B. Early visual selectivity as a function of pattern variables, previous exposure. age from birth and conception. and expected cognitive deficit. In L. B. Cohen and P. Salapatek (Eds.), infant perception: From sensation to cognition. Vol. I: Busic ris~ral proce.tses. New York: Academic Press, 1975. Fantz. R. L.. & Nevis. S. Pattern preferences and perceptual-cognitive development in early infancy. Merrill-Palmer Quarterly. 1%7. 13, 77- 108. Fantz. R. L.. Ordy. J. M., & Udelf. M. S. Maturation of pattern vision in infants during the first six months. Jorrrnal of’ Comparutive rind Ptz~siologicul P.cycholo~~~. 1962. 55, (x17-917. Gaskill. J. D. Lineur systems. Fourier tran.vjhrms. und optics. New York: Wiley. 1978. Gorman, J. J., Cogan. D. G.. & Gellis. S. S. An apparatus for grading the visual acuity of infants on the basis of opticokinetic nystagmus. Pediatrics, 1957. 19, 1088-1092. Gorman. J. J.. Cogan. D. G.. & Gellis. S. S. A device for testing visual acuity in infants.

Si,qht Sa\,ing Review, 1959, 29, 80-84. Greenberg.

D. J.. & O’Donnell,

Development.

W. J. Infancy

and the optimal

level

of stimulation.

Child

1972. 43, 639-645.

Gwiazda, J.. Brill. S.. Mohindra. I.. & Held. R. Infant visual acuity and its meridional variation. Vision Research, 1978. 18, I5577 1564. Haith. M. M. Visual competence in early infancy. In R. Held. H. Leibowitz, & H. L. Teuber (Eds.), Hundbook of Sensory Physiology. Berlin: Springer-Verlag. 1978. Vol. VIII. Harter, M. R.. & Suitt, C. D. Visually-evoked cortical responses and pattern vision in the infant: A longitudinal study. P.+zcmomic Science, 1970. 18, 235-137. Haynes. H.. White. B. L., & Held. R. Visual accommodation in human infants. Science. 196.5. 148, 528-530. Hershenson. M. Visual discrimination in the human newborn. Journul oj’Compurati~~e and Phy~iologicul Psychology. 1964, 58. 270-276. Hershenson. M.. Munsinger. H.. & Kessen, W. Preferences for shapes of intermediate variability in the newborn human. Science. 1%5. 147, 630-631. Hilz. R.. & Cavonius. C. R. Functional organization of the peripheral retina: Sensitivity to periodic stimuli. Visual Reseurch. 1974. 14, 1333- 1338. Hoekstra. J.. Van der Goot. D. P. J.. Van den Brink, G.. & Bilson. F. A. The influency of number of cycles upon the visual contrast threshold for spatial sine patterns. Vision Rereurch, 1974. 14, 365-368. Kagan. J.. The determinants of attention in the infant. American Scientist, 1970, 58, 298-305. Karmel. B. Z. The effect of age. complexity, and amount of contour on pattern preferences in human infants. Journul of ExperimentuI Child Psychology. 1969. 7, 339-354. (at Karmel. B. Z. Complexity, amounts of contour. and visually dependent behavior in hooded

44

BANKS

rats,

domestic

chicks,

Psychology. Karmel,

Basic

visual

P. Salapatek

(Eds. New

processes.

D. H. Spatial frequency D. H. Pattern detection checkerboards D. H..

&

transform:

Circular

Leehey,

S. C., vision.

infant Lewis,

T. L.,

Maurer.

D..

Child

E. M..

selectivity and

&

c$ Compurative

und Phy.~iologicu/

1975.

A., Brill, 190, 9OG903.

dt Kay.

D. Newborn’s

visual

attention.

and 1975.

S.,

Reseurch,

1975, transform:

Fourier

Research,

&

Vol.

I:

l&665-67:. Flickering

1976. 16. two-dimensional

the

In L. B.

ro c~r~~nitir~/>,

177-287. Fourier

15, 91 I-915.

Held,

central

R.

Orientational

vision:

anisotropy

Whole

in

JorrrnuI

or hole?

I?/

1978, 26, 193-203.

Peltzman,

Evoked

infant

Vision

detection

Research.

of /he Optical

D. N..

infants:

for

Frctnr sensatictn Press. 1975.

Vision

mechanisms. S. Pattern

J. J. Sine-wave

Journal

model

in the retina. two-dimensional

the

Psychology.

DePalma,

activity

Vision

targets.

Moskowitz-Cook, Science.

phenomenon. Marg. E.. Freeman. human

JOUYIIUI

1. fn&mr prrce&n: York: Academic

and chromatic Magnuski, H.

Experimrrrtul Lowry.

infants.

(b) E. B. A neuronal

B. Z.. & Maisel. and

Kelly.

human

SALAPATEK

1969, 69, 649-657.

Cohen Kelly. Kelly,

and

AND

potential

response

in the

visual

system,

I.

The

Mach

Society

of‘ America, 1961, 51, 740-746. P.. & Goldstein. P. J. Visual acuity development

Investi.~utive

measurements.

in

Op1zthulmolog~.

1976.

15, 150-153. McCall.

R. B.. & Melson,

W. H. Complexity.

McCarvill, lished McCarvill, Infant 374.

area

University of Connecticut. B. Z. A neural activity

thesis. & Karmel.

Journal

preferences.

A. E. Infant’s

discrimination

Child

S. Visual

Psychology.

abilities

and

1973. interpretation

c$ Experimental of internal 1976. 22,

pattern

of

by Ratliff.

M., Spinelli, D.. evoked potentials. F. Mach

A.,

& Maffei.

L.

Brain Research, 1978. 141, Quanriratib’e studie.\ on neural

hands:

cisco: Holden-Day. Robson, J. G. Spatial

Journal

Fiorentini.

1965. and temporal

of’ the Optical

Society

pattern

Unpuheffects

1976.

on

22,

363-

Journul

elements.

premature

infants

and

of

full-term

1970. 10. 189-205. discrimination

in two-month-old

Infant contrast 179-184.

nrfrwrL>

contrast-sensitivity

of’ America.

irr infirnts.

Psycho/ogy,

and external 129-246.

preferences

of attention

of luminance

Child

neonates. Joctrnul of Experimentul Child Psychology, Peeples, D. R.. &Teller, D. Y. Color vision and brightness human infants. Science, 1975, 189, 1102-l 103. Pirchio,

as determinants

Master’s S. L.,

Experimental Miranda.

and

Psychology. 1970. 3, 343-349. oj’srimulus intensity on puttern pryfkrrnces

pattern

Milewski.

contour

Developmental S. L. The ejrecrs

in infants.

sensitivity

evaluated

in the rrtinu.

functions

of

the

visual

San

Frarlsystem.

1966, 56, I l41-

I 142. ofconcentricity.

Ruff.

H. A.. & Birch, H. G. Infant visual fixation: The effect curvilinearlry. and the number of directions. Jownul of’E.rperimenru/ Ps~c~hok~,y~. 1974. 17, 460-473. Salapatek, P., & Banks, M. S. Infant sensory assessment: Vision. In F. D. Minifie and L.. I... Lloyd (Eds.), Communicarive and cognitive ahilifies: Euriy beharaiorul ussessmcn~. Baltimore: University Park Press, 1978. Salapatek, P., Bechtold. A. G.. & Bushnell. viewing distance. Child Development, Savoy. R. L.. & McCann, J. J. Visibility

E. W. 1976, 47, of low

Infant X60-863. spatial

visual frequency

acuity

as a function sinewave

Dependence on number of cycles. Journul ot’the Optical Society of Americu. 343-350. Schade, 0. H. Optical and photoelectric analog of the eye. Jvurnul ctf rhc, Opticul America, 1956. 46, 721-739. Sokol. S. Measurement of infant visual acuity from pattern reversal evoked Vision Research, 1978, IS, 33-40.

ot

targets:

1975. 65. Sociefy potentials.

o/

INFANT

Sokol,

S., & Dobson,

V. Pattern

Ophthalmology, Stirnimann.

F. Ueber

D. Y.,

Teller,

1976.

van

gratings Meeteren,

Van

TN0 Nes,

Report F. L..

Hurvich Wilson.

visually

evoked

Farbenempfinden

R..

Borton,

R.,

No. IZF-1966-7, & Bouman, M.

transfer. G. Visual

Excerpta acuity

& Regal.

& Giese.

Medica

and

Handbook

(Eds.), VII. H. R.,

Investigative

in infants.

Anna/es

D. Visual

spatial

effects

Paediatrici,

acuity

for

of wavelength

International

of sensoy

S. C. Threshold

thresholds.

physiology.

and

New

E. B., & Karmel, Paper presented

infants.

Philadelphia, A. F.,

presented 15-18.

RECEIVED:

March, & Banks,

at the

Society

B. 2. Failure at the meeting

luminance

Academic

gradient

Perception on 125. and

Press,

visual 1965. L. M. 1972.

Vision

patterns.

1970.

NOTES

to replicate of the

the bull’s

Society

for

eye preference

Research

in Child

effect Develop-

1973. M. for

S. The human Research

in

face: Child

A view from Development,

the infant’s San

Francisco,

1979.

November

diagonal

Springer-Verlag.

of frequency

York:

for

Serial No. In D. Jameson

Heidelberg:

visibility

1944. 163,

vertical

and

Congress,

modulation

REFERENCE Maisel,

ment, 2. Souther,

potentials

Neugeborener.

1967. A. The

Research, 1977. 17, 1177-l 190. Zusne. L. Visual perception ofform.

I.

45

in human infants. Vision Research, 1974. 14, 1433-1439. A. Spatial sinewjave response of the visual system. Institute

modulation Westheimer, Vol.

reversal

VISION

15, 58-62.

das

Morse,

PATTERN

30. 1978;

REVISED:

June

8, 1979,

October

10.

1979.

eve.

Paper March

in