Detection and direction discrimination performance with flicker gratings in peripheral vision

VisionRes.Vol. 34, No. 6, pp. 763-713. 1994

Copyright Q 1994 Elsevier Science Ltd Printed in Great Britain. All rights reserved 0042-6989/94 $6.00 f 0.00

Detection and Direction Discrimination Performance with Flicker Gratings in Yeripheral Vision JAMES MCCARTHY,*?

ALLAN PANTLE,*$

ALAN PINKUS$

Received 25 January 1993; in revisedform 30 June 1993

Detection and direction discrhnination experiments were couducted with luminance and fiicker gratings. The flicker gratings had bars made up of static random pixels interspersed between other bars with fiickering random pixels. AU experiments were carried out in peripheral vision with grating images centered at tdeg eccentricity in the superior retina. Detection of tlicker gratiugs (i) was independent of pixel size, (ii) declined with spatial frequency in the range l-4 cfdeg, and (iii) improved with stimulus area (number of grating cycles). Detection performance with a flicker grating was comparable to that obtained with a low-contrast (0.01) luminance grating, and the results suggest that the spatial structure of a ~icker~om~n stimulus is based upon signals which are weak compared to the maximum signals attainable with a luminance-domain stimulus. With the detectability of flicker and luminance gratings equated, d’ for discriminating the direction of motion of a luminance grating increased with step size (l/12 to l/4 cycle) whereas direction nomination ~~0~~ with a fiicker grating remained unchanged and at chance levels. Under the coudiions tested, there was no evidence that the motion of a ffickerdomain stimulus could be processed peripherally. Constraints on alternative models of motion processing are discussed. Apparent motion

Peripheral vision Flicker gratings Second-order stimuli Direction discrimination

The latter comparison bears on models for the processing of first- and second-order motion (e.g. Petersik, 1989; Cavanagh & Mather, 1989; Wilson, Ferrera & Yo, 1992). A specific working model is presented in the introduction to Expt 2 after luminance and flickerdomain stimuli are described (next section) and their detectabilities equated (Expt 1).

INTRODUCTION Different attributes or dimensions can provide the basis for the definition of the spatial structure of a stimulus. A substantial amount of theoretical work hinges on a classification of stimuli defined by luminance (termed luminance-domain or first-order stimuli) and stimuli defined by other attributes (nonIuminance-domain or second-order stimuli) (e.g. Bergen & Landy, 1991; Cavanagh & Mather, 1989; Chubb & Sperling, 1988; Pantle, 1992). One example of a second-order stimulus is a flicker-defined stimulus (Sperling, 1976; Petersik, Hicks & Pantle, 1978; Prazdny, 1986). The contrast of dots in some region(s) of a random-dot pattern is reversed at one temporal rate while that in other regions is static or reversed at a different rate. In the present expe~ments we investigate some factors which influence the detection of a grating whose alternate bars contain static and flickering pixels. In addition, we compare directly the abilities of a luminance and a flicker-domain grating to support a motion percept in peripheral vision. *Departmentof Psychology,Miami Oxford, OH 45056, U.S.A.

University,

104 Benton

Hall,

tPresent address: Sonalysts Inc., 215 Parkway North, P.O. Box 280, Waterford, CT 06385, U.S.A. $To whom all correspondence should be addressed. 8Armstrong Laboratories, Wright-Patterson Air Force Base, OH 45433, U.S.A.

GENERALMETHOD The experiments were conducted with flicker-domain and luminance-domain stimuli. Flicker-domain stimuli were the main focus of the experiments, and they are described first. Flicker-domain stimuli A flicker-domain stimulus was a multiple-frame sequence which, for ease of description, can be broken down into four blocks of frames. Each block contained a pair of frames, A and B, which differed from one block to the next. Frame A of each pair was an independently generated random dot-pattern. It was made up of an N(rows) x A4(columns) array of light (9.4 mL) and dark (0.9 mL) pixels. There was an equal probability that a pixel was light or dark. Two different types of patterns were used for Frame B. The different types were defined by their relationship to Frame A. For a ~t~uctu~e~ flicker-domain stimulus (a flicker

763

764

JAMES

MCCARTHY

grating), Frame B was composed of a number of pixel sets. Each set spanned the entire horizontal width of the stimulus, but only a fraction of its total height. Half of the sets were exact negatives of the pixels occupying corresponding locations in Frame A. The remaining sets of pixels of Frame B were identical to (positive versions of) the corresponding regions of Frame A. The positive pixel sets were interleaved spatially (in the vertical direction) between the negative pixel sets. When Frame A and Frame B were alternated in time, an observer perceived a horizontal square-wave flicker grating. Each spatial cycle of the flicker grating had one half cycle composed of a set of flickering pixels (a dynamic bar) and a half cycle of a set of steady pixels (a static bar). The point-by-point luminance correlation between Frames A and B was - 1.O for the flickering regions and + 1.O for the static regions. For a non -structured flicker-domain stimulus, the pixels of Frame B were generated independently of those in Frame A. In other words, the point-by-point luminance correlation between Frame A and Frame B was, on average, 0.0. As with the structured stimuli, half of the pi::els of an entire frame reversed contrast each time the stimulus switched from Frame A to Frame B or vice versa, but the reversing pixels were distributed randomly across the entire frame. The non-structured stimuli were used as “noise” stimuli in the detection tasks (described later) of Expts IA and 3-5. For both structured and non-structured stimuli, each frame (A and B) in a given frame block lasted for two screen sweeps (33 msec total), producing an A-B frame frequency of 15 Hz. The pair of frames was alternated three times per frame block, making the duration of a frame block 198 msec and the total duration of a stimulus sequence 792 msec for a four-block sequence. In motion discrimination tasks (described later) with flicker gratings (Expt 2) the new A-B pair used for each frame block produced a flicker grating in a new position, and it resulted in apparent motion of the grating. Also, the use of different A-B pairs in each frame block meant that there was no coherent motion of pixels across frame blocks. In detection tasks with structured and non-structured stimuli (Expts IA and 3-5), new A-B pairs were used for each frame block to prevent subjects from making discriminations about structure on the basis of any fortuitous, salient groups of pixels. In each experiment eight different stimulus sequences (four up and four down motion sequences in Expt 2, and four structured and four unstructured sequences in Expts 1A and 3-5) were employed in order to ensure that discriminations were not due to the peculiarities of any one sequence. The flicker-domain stimuli were displayed on a Zenith monochrome monitor controlled by a 6809-based microcomputer. Stimulus frames were generated in advance and stored in the computer. The frames were displayed at a non-interlaced refresh rate of 60 Hz. The monitor had a P31 phosphor. Stimuli were centred behind and seen through a rectangular hole in a white matte surround. The average luminance of the stimuli was

rl al.

5.15 mL. The pixel contrast was 83%. The luminance of the surround was 4.1 mL, and it contained fixation marks located at various distances from its center. Luminance-domain stimuli

Analogous to the structured and non-structured flicker-domain stimuli, there were two luminancedomain stimuli-a luminance grating with a squarewave profile and a spatially uniform field. The space-average luminance of the gratings and the uniform field was 5.7 mL. The contrast and spatial frequency of the grating could be set to various values required by the experiments. In motion discrimination tasks with the luminance grating (Expt 2) its position was updated every 198 msec to produce apparent motion with the same temporal characteristics as was used for the flicker grating. The luminance-domain stimuli were presented on the face of a Hewlett-Packard display scope (Model 1332A) coated with a P31 phosphor. The electronic signals which produced the stimuli were generated by a hybrid system of analog equipment and a Z-80 based microcomputer. The beam of the CRT was swept vertically by a ramp signal of 222 Hz and horizontally by a freerunning triangular-wave signal of 1.4 MHz. The vertical sweep was triggered by the computer, and in synchrony with each sweep, a series of intensity values was read from a table stored in the computer and applied, via a digital-to-analog converter, to the Z-axis of the scope. The net result was that a stimulus pattern made up of 256 horizontal strips, each at one of 256 possible gray levels, was painted in succession on the CRT within 4.5 msec. Subjects viewed the CRT through a rectangular hole cut in a white matte piece of cardboard used as a surround. The luminance of the surround was 3.1 mL. Psychophysical tasks D ’ measures of sensitivity were obtained for two different types of psychophysical judgments. For a detection task subjects decided whether a stimulus possessed spatial structure or not. In the flicker domain this amounted to a discrimination of a flicker grating (structured stimulus) from a uniformly flickering field (nonstructured stimulus) of random dots. In the luminance domain, a subject was required to discriminate the square-wave luminance grating (structured stimulus) from a spatially uniform field (non-structured stimulus). If the subject reported that he/she saw a grating, when in fact the grating was presented, the response was recorded as a “hit”. If he/she reported that a grating was present, when in fact the uniformly flickering field (flicker domain) or the spatially uniform field (luminance domain) was presented (noise trial), the response was scored as a “false alarm”. D’ was computed from the “hit” and “false alarm” rates and was used as a measure of a subject’s sensitivity to a grating. The d’ for detection of grating structure in each condition of all experiments was based on 80 trials, 40 grating trials and 40 noise trials. For the flicker-domain stimuli the 80 trials were distributed over two sessions of 40 trials each. For the

FLICKER GRATINGS IN PERIPHERAL

luminance-domain stimuli all 80 trials were completed in one session, For both types of stimuli, half of the trials within a session were grating trials; the other half, noise trials. The order of grating and noise trials was random. Logically, the flicker-domain stimulus used on noise trials could have taken many forms. For example, a totally static field of random dots, a pattern of randomly scattered regions of flickering dots, or a random-dot field with all dots flickering could have been chosen. The stimulus sequence with independently generated Frames A and B was chosen for two reasons. (1) As noted earlier, with this type of unstructured stimulus the same number of pixels (50%) change from frame to frame as change in the structured sequences (flicker gratings). (2) Pilot work indicated that the flicker gratings were indistinguishable from the chosen noise stimulus when their structure was not visible. The second type of psychophysical task was a direction discrimination judgment. A subject decided whether a stimulus with structure (flicker or square-wave luminance grating) moved up or down. There was an equal probability that the grating actually moved up or down. If the subject reported that the grating moved up, when in fact it did, the response was considered a “hit”. If the subject reported that the grating moved up, when in fact it moved down, the response was scored as a “false alarm”. “Hits” and “false alarms” were used to compute a d’ measure of direction sensitivity. As with the detection task, the d’ for direction dis~imination for each condition of all experiments was based on 80 trials. For 40 trials the grating moved up, and for the other 40 trials it moved down. For flicker- and luminance-domain stimuli the number of sessions, trials per session, and the order of presentation of trials within a session followed the same formats as those used for the detection task. For both detection and direction discrimination tasks, a subject fixated a point marked on the surround for each stimulus in order to place the center of the stimulus image 8 deg in the superior retina. In a similar study with contrast-modulated gratings, Pantle (1989) measured eye movements and found that subjects were able to maintain a steady fixation for this eccentricity. Subjects participated in a training session in which they were given a chance to view structured and non-structured stimuli and to see the structured stimuli move. Otherwise, subjects were not given feedback about their performance. EXPERIMENT1 The first experiment had a twofold purpose. (1) It was one in a series of four experiments in which we used signal detection procedures to quantify the detectability of flicker gratings. (2) Using signal detection procedures, we wanted to determine the contrast of a luminance grating which made it just as detectable in peripheral vision as a flicker grating with matched characteristics (spatial frequency, size, etc.). With det~tabiliti~ equated, motion sensitivity for luminance and flicker gratings was compared in Expt 2.

765

VISION

TABLE 1. Visual angies of the fecttures of the stimulus displays at the various viewing distances for Expts IA and 3

Medium Large Viewing Spatial Grating Smll distance frequency size pixelsize pixelsize pixel size (m)

Wdeg)

2.0

1.00 1.25 1.50 1.75

2.5 3.0 3.5

(deg) 4.0 3.2 2.7 2.3

x x x x

4.0 3.2 2.7 2.3

(min) 2.5 x 2.0 x I.7 x 1.4 x

@in)

2.5 2.0 1.7 1.4

5.0 x 4.0 x 3.3 x 2.9 x

5.0 4.0 3.3 2.9

Wn) 10.0 x 8.0 x 6.7 x 5.7 x

10.0 8.0 6.7 5.7

Experiment IA The flicker gratings used in Expt 1A were 96 pixels high by 96 pixels wide. Each flickering and each static bar of a flicker grating was I2 pixels high so that any one frame contained 4 horizontal grating cycles. Viewing distance was varied in the experiment, and the consequences of changes in viewing distance on the visual angles subtended by the various features of the display are summarized in the first four columns of Table 1. Subjects fixated a different point on the surround at each viewing distance to keep the CRT image centered at an eccentricity of 8 deg. Each subject completed the detection task for each viewing distance. The order in which the viewing distances were employed was counterbalanced across subjects. The proportions of “hits” and “false alarms” were used to compute d’s for detection of the flicker grating at each viewing distance for each subject. In order to compute d ’ for the few cases for which the “hit” rate was lOO%, the percentage was arbitrarily reduced to 99.5% as though the subject would have produced one “miss” in 200 trials. Correspondingly, “false alarm” rates of 0% were arbitrarily increased to 0.5% as though the observer would have produced one “correct rejection” in 200 trials. This adjustment of “hit” and “false alarm” rates was employed in all experiments. The pattern of results was essentially identical for all subjects. For this reason d’s averaged across the individual subjects are plotted in Fig. 1 as a function of viewing distance. An analysis of variance of the d’s revealed a statistically significant effect of viewing distance 5

4l-

d’

r

I

3!-‘\

\

2

I

0’

I

I

2.0

2.5

I

t

3.0

3.5

Viewing distance (m) FIGURE 1. D ‘ for the detection of a flicker grating as a function of viewing distance. Each data point is the mean for six observers. Verticalbars represent f 1 SE.

JAMES MCCARTHY et al.

766

[F(3,15) = 92.66, P < O.OOOl]. Mean d’ decreased monotonically with viewing distance, from a high of 3.80 at the smallest distance to a low of 0.38 at the largest distance. The values of d’ indicate that the subjects had little trouble discriminating the flicker grating from the random noise field at the closest viewing distance, but that the grating was virtually indiscriminable from the noise field at the largest distance. Because the change of viewing distance produced concomitant changes in overall stimulus size, grating spatial frequency, and pixel size, the abscissa could represent values for any of those variables as well. Indeed, it is likely that those variables alone or in combination, rather than viewing distance per se, were the critical determinants of the detectability of the flicker grating. The individual contributions of the different variables were examined further in Expts 3-5. Aside from looking at the possible effect of viewing distance and spatial variables on the visibility of the flicker grating, the second goal of Expt 1 was to quantify the detectability of flicker gratings so that their detectability could be equated with that of luminance gratings. In order to equate the detectabilities, some criterion level had to be chosen. A d’ of 3 was selected. After fitting a regression line to the data of Fig. 1 (d’ = 2.35 x viewing distance + 8.35) by the Method of Least Squares, the equation was solved for viewing distance with d’ set to 3. The result was a viewing distance of 2.3 m. At this distance the overall size of the flicker grating was 3.5 deg high by 3.5 deg wide, and the spatial frequency of its bars was 1.13 c/deg. Virtually identical parameters were used for the luminance grating in Expt 1B. Experiment

1B

Having determined the stimulus parameters which produced a d’ sensitivity of 3 for the flicker grating, it remained to determine the contrast of a luminance grating which resulted in the same criterion level of discrimination. In Expt 1B the same six subjects who served in Expt 1A were required to discriminate a square-wave luminance grating from a uniform field of the same space-average intensity. The grating was 3.5 deg high by 4.4 deg wide at a viewing distance of 1.6 m. Its spatial frequency was 1.13 c/deg. Its contrast was set to one of five different values--0.003, 0.006, 0.009, 0.012 or 0.015. Only one contrast value was used per session, and the order of presentation of the contrast levels was counterbalanced across subjects. The proportions of “hits” and “false alarms” were used to compute d’s for the detection of the luminance grating at each contrast for each subject. The pattern of results was identical for the six subjects, and the mean d’ at each contrast is shown in Fig. 2. D ’ was a positive function of grating contrast, and the effect of contrast was statistically significant [F(4,20) = 39.74, P < O.OOOl].At the lowest grating contrast (0.003), the d’ of 0.83 indicates that the grating was nearly indiscriminable from the uniform field. At the highest contrast

3 i’

d’ 2

o(

$

0.003

/

0.006

0.009

0.012

0.015

Contrast FIGURE 2. D 'for the detection of a square-wave luminance grating of 1.13 c/deg as a function of grating contrast. Each data point is the mean for six observers. Vertical bars represent f 1SE.

(0.015), the grating was almost perfectly discriminable (d’ = 4.22) from the uniform field. A regression line was fit to the contrast data by the Method of Least Squares: d’ = 284.67 x contrast + 0.088. Using the regression equation, it was determined that a contrast of 0.01 would yield a d’ sensitivity of 3. What this means is that with respect to the discriminability of structure, a luminance grating had to possess a contrast of 0.01 to be equivalent to the flicker grating. Finally, supplementary measurements were made to ensure that the equivalent-contrast value of 0.01 held for the apparently moving gratings of Expt 2. In Expt 2 sensitivity to the motion of a flicker grating was compared to that of an equally detectable luminance grating. It should be noted that 0.01 was an estimate of the contrast for which a d’ of 3 would be achieved by the subjects as a group because the estimate was based upon group d’s in Fig. 2. For a contrast of 0.01, the range of expected d’s for the detectability of the luminance grating by individual subjects was 2.55-3.75. Similarly, the 2.3 m viewing distance which was derived earlier and which was expected to yield of d’ of 3 for the detectability of flicker gratings was a group estimate. The expected range of d’s for individual subjects was 2.39-3.70. That the individual differences were ignored safely in equating detectabilities was borne out by the fact that there were no statistically significant correlations (P’s > 0.05) between the individual expected detectabilities of gratings, either luminance or flicker, and direction discrimination performance in Expt 2. The small individual differences were apparently random, and consequently, not correlated with individual discrimination performance in Expt 2. EXPERIMENT 2 The outcome of a comparison between motion sensitivity to a flicker grating and motion sensitivity to a luminance grating depends upon the manner in which motion is encoded. One reasonable working model incorporates at least partially separate pathways for

FLICKER GRATINGS IN PERIPHERAL VISION

processing first- and second-order stimuli (see Fig. 3). Motion perception which arises from a luminancedomain stimulus has been termed first-order motion. The perception of motion of a flicker grating is just one example of second-order motion. First-order motion processing is represented by the channel on the left in Fig. 3. The stepping, one-dimensional luminance grating, denoted LG (x, t ), is transformed by a spatial array of linear spatio-temporal filters into an output denoted T-LG (x, t ). T-LG (x, t ) is an activity waveform which is dis~buted across the spatial array of filter elements and which shifts across the array when the stimulus changes position. The spatial waveform corresponds to the spatial structure of the luminance grating, and it is assumed that the amplitude of the waveform determines the detectability of the luminance grating as measured in Expt 1B. It follows that there is a specific amplitude [A (T-LG ) = K] which corresponds to a level of detectability d’ = 3. T-LG with amplitude K then enters into a motion energy computation, the details of which are not important here and can be found in a variety of references (e.g. van Santen & Sperling, 1984; Adelson & Bergen, 1985; Heeger, 1992). The output of the motion energy stage is a direction-opponent signal whose magnitude is assumed to be directly related to a subject’s direction discrimination performante-i.e., a subject’s ability to discriminate the direction of motion of the luminance grating. As others have suggested, the representation of the spatial structure of second-order stimuli (e.g. Bergen & Landy, 1991) and the generation of second-order motion signals (e.g. Chubb & Sperling, 1988) is more complex and requires nonlinear transformation stages. A channel adequate for the processing of flicker gratings is shown on the right in Fig. 3. The luminance at a point (x,y) at time t of a stepping flicker grating is denoted FG (x, y, t ). The functional notation for the flicker grating contains two spatial parameters (x,y) because the spatial luminance distribution of the flicker grating, unlike the luminance grating, is two-dimensional. However, the flicker grating can be considered to be spatially one-dimensional [FG (x, t )] and analyzed

I

Nonlinear transform t

FIGURE 3. Working model of the generation of motion signals by luminance- and ~cker-do~in stimuli. See text for details.

167

accordingly because flicker rate, which defines the grating, varies in a direction perpendicular to the bars, but not parallel to them. The Bicker grating is transformed in three stages to an activity waveform that corresponds to its perceived spatial structure. The first stage is a spatio-temporal filter tuned to the dominant spatial frequency of the grating pixels and differentially sensitive to the flicker rates of the static and flickering pixels. Due to the temporal filtering, the amplitude of variation of the outputs of the filters located in static and flickering regions of the grating would be different, as well as the temporal rate of modulation. After the next transformation, a pointwise nonlinear transformation such as rectification, there would be a difference of the average output for elements corresponding to static and flickering regions of the grating. A smoothly varying activity waveform [T-FG (x, t )] would be present at the outputs of a spatial array of linear spatio-temporal filters at the next stage, and the waveform would shift across the array when the flicker grating moved. Analogous to T-LG, it is assumed that the amplitude of the activity waveform [A (T-FG )] determines the ~tecta~i~ify of the flicker grating, and that there is a specific amplitude [_4(T-FG ) = C] which corresponds to the level of detectability d’ = 3. T-FG then enters into the same motion energy computation as T-LG. For the present purposes, the exact rule for combining the inputs T-LG and T-FG at the motion energy stage is not important because none of the experiments employed superimposed luminance and flicker gratings, but some form of linear summing on “OR” (Cavanagh, Arguin & von Grunau, 1989) function would be adequate. In the model the empirical procedure of equating the det~tabilities of the luminance and flicker gratings at a level d’ = 3 is equivalent to setting K = C. Assuming that the inputs T-LG and T-FG are equally weighted at the motion energy stage, direction di~~mination performance with luminance and flicker gratings ought to be equivalent when K = C, that is, when the detectabilities of luminance and flicker gratings are equated (d’ = 3). This is the primary hypothesis tested in Expt 2 with stimuli located at an eccentricity of 8 deg. There is some evidence that the equal-performance hypothesis holds for centrally viewed luminance and flicker gratings, as well as for centrally viewed luminance and amplitude-modulated gratings (Pantle & McCarthy, 1988). However, Pantle and McCarthy (1988) and Pantle (1992) reported that motion sensitivity to a peripherally viewed flicker grating did not compare favorably with that of motion sensitivity to a luminance grating. A difficulty with the earlier studies with flicker gratings was that there was no guarantee that the detectabilities of the flicker and luminance gratings were not the source of the observed differences in motion sensitivity. when K is not equal to C in the model, it is not expected that direction discrimination with luminance and flicker gratings would be comparable. Six subjects (those from Expts 1A and 1B) attempted to discriminate the upward and the downward movement of flicker and luminance gratings. For both types

768

JAMES MCCARTHY

*r

d’

O-1

I. I

--_.

_--

-

r--A

I

I

I

I

I

2124

3124

4124

5124

6124

Jump size (cycles) FIGURE 4. D ’ for the discrimination of the direction of motion of a luminance grating (solid line) and a flicker grating (dashed line) as a function of jump size. Each data point is the mean for six observers. Vertical bars represent k 1SE.

of stimuli the motion was discontinuous, as described in the General Methods section. In different experimental conditions the step size between successive positions of the gratings was l/12, l/8, l/6, or l/4 cycle. There were three jumps (four grating phases) per motion sequence and the sequence lasted 792msec. The flicker grating subtended a visual angle of 3.5 deg vertically by 3.5 deg horizontally. It was 96 pixels high and 96 pixels wide, and each grating cycle (a pair of horizontal flickering and static bars) was 24 pixels high (a spatial frequency of 1.13 c/deg). The luminance grating subtended a visual angle of 3.5 deg vertically by 4.4 deg horizontally. The spatial frequency of its horizontal dark/light bars was 1.13 c/deg, and their contrast was 0.009 (slightly less than the 0.01 contrast needed to make the luminance grating as detectable as the flicker grating). In order to prevent subjects from relying solely on either initial or final position of the gratings as a cue for direction of movement, ‘two types of up and down motion sequences were used. The gratings either started in the same position and moved to different final positions, or they started in different initial positions and moved to identical final positions. After viewing a number of up and down practice trials with flicker and luminance gratings, subjects completed the direction discrimination task with the flicker and luminance gratings for each jump size. Only one jump size was used per session, and the order of jump sizes was counterbalanced across subjects. Measures of d’ for individual subjects were computed from the proportions of “hits” (upward response to an upward moving stimulus) and “false alarms” (upward response to a downward moving stimulus) for each grating type and jump size. The pattern of results was essentially identical for all subjects, and mean d's are plotted in Fig. 4. The solid function shows d’ for direction discrimination as a function of the jump size for the luminance grating. An analysis of variance of the luminance grating data revealed a statistically significant effect of jump size [F(3,15) = 21.4, P < O.OOOl].As the jump size increased from l/12 cycle to l/4 cycle, d’

et al.

increased from 1.17 to 3.07. For the l/4 cycle shift, the direction of motion of the luminance grating was as discriminable (d’ = 3.07) as was its structure (d’ = 3) in Expt 1B. The results for the flicker grating, shown by the dashed function in Fig. 4, were quite different. Regardless of jump size d’ was near zero, and an analysis of variance revealed no effect of jump size [F(3,15) = 0.5 1, P > 0.681. The near zero d’s for direction discrimination corroborate the subjects’ reports that they were able to see the structure of the grating, but not its motion. The results of Expt 2 agree substantially, but not exactly, with those of Pantle (1992) who also employed peripherally viewed stimuli. Similarities and differences are explored later in the General Discussion. However, the results and implications are different than those obtained with centrally viewed stimuli. Whereas the matching of stimulus detectability across stimulus domains rendered velocity discriminations equivalent in the fovea (Turano & Pantle, 1989). similar matching in the present experiment failed to produce comparable levels of direction sensitivity in the periphery. EXPERIMENT 3 In Expt IA it was not possible to evaluate the individual contributions of overall grating size, grating spatial frequency, or pixel size to the detection of grating structure. Experiment 3 repeated the detection task of Expt 1A with the important addition of an independent manipulation of pixel size which permitted us to dissociate its effects from those of grating size and spatial frequency. Displays with three different pixel sizes were used at each of the viewing distances of Expt 1A. Flicker gratings and noise stimuli whose pixel size matched that of Expt 1A are hereafter termed small-pixel displays. As in Expt 1A, the small-pixel displays were 96 x 96 arrays, and each cycle of a flicker grating was 24 pixels high. The pixels in medium-pixel (large-pixel) displays were two (four) times larger than the pixels in the small-pixel display. The medium-pixel (large-pixel) stimuli were 48 x 48 (24 x 24) arrays with 12 (6) pixels per grating cycle. The visual angles subtended by different features of the three displays at different viewing distances are summarized in Table 1. Two points are worth noting. First, at a given viewing distance only pixel size changes. This manipulation permitted an independent evaluation of pixel size on the detectability of flicker gratings. Second, the range of visual angles (factor of 4) subtended by pixels of different size at each viewing distance is greater than that (factor of 1.75) of a given size pixel at different viewing distances. This relationship is helpful for dissociating the influence of pixel size, grating frequency and stimulus size on detection with changes of viewing distance. The stimuli were centered at an eccentricity of 8 deg in the superior retina. Five new subjects completed the detection task for each of the 12 conditions formed by the combinations of the four viewing distances and the three types of display with pixels of different sizes. The order of presentation

FLICKER GRATINGS IN PERIPHERAL

of the 12 conditions was counterbalanced across subjects. The pattern of d’s for the 12 conditions was the same for all subjects, and mean d’s are shown in Fig. 5. The det~tability of the grating dropped signi~~antly as a distance [F(3,12) = 53.54, function of viewing P < O.OOOl].The effect of viewing distance cannot be attributed to the decreased size of the visual angle subtended by the pixels which accompanied the increase of viewing distance. Over the 1.75-factor increase of viewing distance, the mean d’ dropped from values in the neighborhood of 4 to values near 1. Contrariwise, mean d’s for small-, medium- and large-pixel (a fourfold change of pixel size) displays at any given viewing distance were approximately equal, with a small tendency for the flicker grating to be more detectable with decreased pixel size [F(Z,S) = 4.97, P < O&l]. The interaction of pixel size with viewing distance [F(6,24) = 0.48, P > O.Sl] was not statistically significant. The lack of an interaction means that, at each viewing distance, the relative ability of each size of pixel to serve as the base or carrier for the temporal modulation which defined the difference between the dynamic and static bars of the flicker grating remained unchanged. The remaining two experiments were attempts to determine what other factors might explain the effect of viewing distance. In the next experiment, we isolated and examined the effect of spatial frequency on detection performance with flicker gratings.

EXPERIMENT

4

If spatial frequency were the source of the viewing distance effect in Expts 1A and 3, the decline would correspond to the range l-l .75 c/deg (see Table 1). In the present experiment, we investigated this cutoff region by measuring a subject’s ability to detect flicker gratings of spatial frequencies 1, 2 and 4 c/deg. Pixel size (small, 2.5 min square) and viewing distance (2 m) were held constant. Flicker gratings and noise fields were 96 x 96 arrays of random pixels. The 1, 2 and 4 c/deg gratings were defined by 24, 12, and 6 pixels Per grating cycle,

D

Small

m

Medium

m

Large

769

VISION

I

I

1

2

FIGURE 6. D ’ for the detection of a flicker grating as a function of spatial frequency. Each data point is the mean for four observers. Vertical bars represent + I SE.

respectively. The 1, 2 and 4cfdeg gratings contained 4, 8 and 16 total cycles, respectively. The subjects were four of the five individuals who participated in Expt 3. Each subject completed six sessions, two for each of the three spatial frequency conditions. The order of presentation of the spatial frequency conditions was counterbalance across subjects. Mean d’ values are plotted in Fig. 6 as a function of the spatial frequency of the flicker grating. The overall decline of d ’ with spatial frequency is statistically significant [F(2,6) = 155.64, P < O.OOOl].Detection performance decreased slightly between 1 and 2 c/deg and more precipitously between 2 and 4 c/deg. The change of d’ between 1 and 2 c/deg (about 1.5) is small compared with the change of d ’ obtained with the variation of viewing distance in Expts 1A and 3. In those experiments the increase of viewing distance from the nearest to the farthest point would have increased the spatial frequency of the flicker grating from 1 to 1.75 c/deg. Some additional factor must have been responsible for the larger decrease of detection performance in Expts 1A and 3. One possibility is the overall size of the stimuli and the number of grating cycles. In the present experiment the number of cycles increased with spatial frequency as grating size was held constant, whereas the number of grating cycles remained fixed in Expts IA and 3 as grating size decreased with increased viewing distance. The effects of grating size and number of grating cycles on the visibility of flicker gratings were explored in the next experiment. EXPERIMENT

Viewing distance (m) FIGURE 5. D ’ for the detection of a Bicker grating as a function of viewing distance. Results for three displays with different pixel sizes are shown. Bar height is the mean for five observers. Vertical bars represent + 1SE.

4

Spatial frequency (cldeg)

5

Experiment 5 was a set of supplementary measurements of detection behavior which could be compared directly with the performance of the same subjects in the 2 c/deg condition of Expt 4. The supplementary condition differed from that of the 2 c/deg condition of Expt 4 only in that they were made with a smaller window which reduced the overall size of the stimuli from 96 x 96 pixel arrays to 48 x 48 arrays and the total

770

JAMES MCCARTHY er al.

number of grating cycles from 8 to 4. Five subjects (four from Expt 4 and one other) provided data for the new condition. The advantage of a larger stimulus size and more grating cycles for detection is shown in Fig. 7. The mean d’ for the larger stimulus was greater than that for the smaller stimulus. The difference between the means for the four subjects who served in Expts 4 and 5 is statistically significant [t (3) = 6.14, P < 0.011. While the results cannot be used to establish the cause of the size effect (e.g. probability summation or facilitatory interactions over space), they nonetheless demonstrate that the steeper high spatial frequency cutoff observed in Expts 1A and 3 compared to Expt 4 was probably the result of changing the window size in Expts 1A and 3. DISCUSSION

Empirical conclusions

The detectability of a flicker grating in the periphery was comparable to that of a luminance grating with 0.01 contrast. Variables affected the detection of its structure in the same way that they influence the detection of a low-contrast luminance grating. Visibility of the flicker grating decreased with spatial frequency with a spatial resolution limit in the neighborhood of 4 c/deg. At 8 deg eccentricity, the resolution limit for a luminance grating is similar (Robson & Graham, 1981). Like luminance gratings (Robson & Graham, 1981) flicker gratings became less detectable when the number of grating cycles was reduced from 8 to 4. In general, the spatial structure of the flicker was visible peripherally, but it appeared to be based on signals which were weak compared to the maximum signals attainable with luminance-domain stimuli.

“r

d’

2

4 cycles

8 cycles

Window size FIGURE 7. D ’ for the detection of a flicker grating as a function of stimulus area (number of grating cycles). Bar height is the mean for five observers. Vertical bars represent k 1 SE.

With detectabilities of luminance and flicker gratings equated in peripheral vision, direction discrimination with a luminance grating increased with step size (l/12 to l/4 cycle), whereas direction discrimination with the flicker grating remained unchanged and at chance levels. In an earlier study Pantle (1992) had found some improvement of direction discrimination with large step sizes of a flicker grating. The bars of the flicker grating employed by Pantle ran parallel to an imaginary line drawn from the fixation point to the grating. The arrangement made it possible to judge changes of the vernier alignment of the grating with respect to the fixation point. With large step sizes there was reason to believe that subjects might have used changes of the grating’s position to infer direction of motion when in fact they saw no actual motion. In the present study the bars of the flicker grating were oriented perpendicular to an imaginary radial line from the fixation point. Vernier alignment of the grating with the fixation point was not possible. With this arrangement, there was no evidence that the motion of the peripherally located flicker grating could be detected either directly or indirectly through position cues. Contrast equivalence Within-class equivalence.

More often than not in vision, the term contrast has been used to refer to a difference between the luminances of the parts of a stimulus. The greater the luminance difference, the greater is the degree of contrast. Because stimuli can be defined by differences other than luminance, the term contrast or modulation is also used to refer to those differences as well. For example, gratings with alternating red and green stripes have color contrast; gratings whose alternate stripes contain static and flickering pixels have flicker contrast; gratings whose alternative stripes contain pixels moving in different directions possess motion contrast; or gratings whose alternate stripes are composed of a low- and high-contrast carrier have contrast modulation. In all cases, contrast refers to a single dimension of difference between stimulus parts. Performance often, but not always, varies with the degree of stimulus contrast in a psychophysical task. For this reason, a decision must be made about what stimulus contrasts to use when comparing performances across stimuli. Within a class of stimuli defined by the same attribute or dimension (e.g. luminance-domain stimuli) it is possible to match physical contrasts or subjective contrasts defined by some psychophysical criterion, such as d ‘ or multiples of the detection threshold. Matching based upon subjective visibility is clearly advantageous when it leads to simpler explanations. For example, in central vision once the effect of exposure duration or retinal illumination on detection thresholds is taken into account, there is little further effect of exposure duration or retinal illumination on vernier thresholds for abutting targets. In other words, variations in vernier acuity can be reduced to variations in detection thresholds. The findings suggest that similar spatial mechanisms mediate detection and vernier

FLICKER GRATINGS IN PERIPHERAL

thresholds and that both thresholds are limited by the same factors (Waugh & Levi, 1993a,b). Failures to achieve equivalent performance in some task for equally detectable stimuli within a single attribute class can be just as informative. In central vision the color contrasts required to perceive the direction of motion of color gratings are comparable to the color contrasts required to detect them (~~ington & Henning, 1993). In parafoveal vision however, direction discrimination thresholds are higher than detection thresholds for color gratings (Lindsey & Teller, 1990; Derrington & Henning, 1993). And, at 8 deg eccentricity, Pantle (1992) found that discrimination of the direction of motion of a color grating was near chance even when the color contrast of the grating was 16 times its detection threshold. For color gratings then, direction discrimination as a function of eccentricity cannot be reduced to differences in color contrast detection thresholds. Other factors must be involved. Pantle (1992) had reached a similar conclusion about contrast modulation based upon detection vs direction discrimination thresholds in central and peripheral vision. Between-class equivalence. Logically, there is no way to match physical contrasts across stimulus classes. For example, what contrast of a luminance grating could be said to match physically the flicker contrast for a grating which contained some stripes with static pixels and some with pixels flickering at 15 Hz? Subjective matches are possible, however, and different researchers have used multiple psychophysical criteria to define such matches. Turano and Pantle (1989) scaled luminance and contrast-modulated gratings in terms of their own respective detection thresholds. When the contrast of the luminance grating and the modulation of the contrastmodulated grating were set at equal multiples of their respective detection thresholds, fovea1 velocity discrimination performance with the two types of gratings was quantitatively the same. This meant, for example, that a contrast-modulated grating with maximum modulation (lOO%, approx, 10 times threshold) was as detectable as a low-contrast (2%, approx. 10 times threshold) luminance grating, and it resulted in the same velocity discrimination performance. Under these circumstances it was suggested that the luminance and flicker gratings provided equal inputs to a common motion mechanism. A similar set of findings was obtained by Stoner and Albright (1992) studying the coherent motion of plaids in central vision. If, among other key parameters, the relative contrast of two superimposed luminance gratings moving in different directions is adjusted properly, subjects will perceive the combination as a coherent plaid pattern moving in a single direction (Adelson & Movshon, 1982). What Stoner and Albright did was to replace one of the luminance gratings with a flicker *Although our model has two separate pathways whose inputs to a motion stage are assumed to be equated, for the experiments reported here the model would be equivalent to one in which the two stimuligeneratedone identical signal to serve as the input for the motion mechanism.

VISION

771

grating. Flicker contrast was defined by the varying probability that individual pixels along the spatial extent of the grating would undergo a contrast reversal across frames. Significantly, Stoner and Albright found (1) that the luminance grating would combine with the flicker grating to form a coherent pattern, but that (2) the highest degree of coherence was obtained when the luminance grating had a low contrast (7%, approx. 3 times detection threshold). Under these circumstances, it was suggested that each grating caused equivalent types of activation in some kind of form-cue invariant, motion mechanism. Stoner and Albright go on to suggest that their “contrast equivalent stimuli might have similar detection thresholds” (p. 473). It is probably theoretically significant that the converse of their hypothesis, namely that stimuli which are defined to be contrast equivalent by virtue of having equal detectabilities, can be equally effective as motion stimuli in different psychophysical tasks when the stimuli are presented foveally (Turano & Pantle, 1989; Derrington & Henning, 1993; Pantle, 1992), but not peripherally (Lindsey & Teller, 1990; Derrington & Henning, 1993; Pantle, 1992; Expt 2). The peripheral results are not consistent with the working model outlined in the introduction to Expt 2, essentially a specific version of a model with a motionsensitive mechanism which exhibits form-cue invariance (Stoner & Albright, 1992). Modifications of the model for peripheral vision are considered in the next session. Alternative schemes for processingjrst motion

- and second-order

The model outlined in Fig. 3 is a specific implementation of a general class of models suggested by a number of researchers (e.g. Turano 8z Pantle, 1989; Cavanagh & Mether, 1989; Cavanagh et al., 1989; Stoner & Albright, 1992; Victor & Conte, 1992). The key feature of the general class of models is the form-cue invariance of a single motion mechanism. For central vision the model is supported by a variety of psychophysical results which demonstrate that the perception of motion is independent of the attributes which define the form or figural properties of a moving stimulus (Turano & Pantle, 1989; Cavanagh et al., 1989; Pantle, 1992; Stoner & Albright, 1992; Derrington & Henning, 1993). The general model also derives support from physiological studies of macaque MT neurons which exhibit form-cue invariance (Albright, 1992), and from human, “motion onset”, evoked potential studies which demonstrate that visual responses (possibly arising as early as Area 17) to first- and second-order stimuli are equivalent over a wide range of velocities and contrasts (Victor & Conte, 1992).* The present results on peripheral vision are in marked contrast to the support for the model in Fig. 3 provided by the aforementioned results for central vision. Equating the detectabilities of the luminance and flicker gratings and their assumed inputs to the motion stage in the model [A (T-LG) = A (T-FG)] did not make their direction of motion equally discriminable. The observed difference in disc~minability contradicts the quantitative

772

JAMES MCCARTHY

prediction of form-cue invariance; i.e., the output of the motion stage should be the same whenever A (T-LG) equals A (T-FG ). Similar quantitative departures from predicted form-cue invariance after equating for detectability have been reported by Pantle (1992) for contrast-modulated gratings and by Lindsey and Teller (1990) and Derrington and Henning (1993) for color gratings in peripheral vision. Failure to find support for the specific model in Fig. 3 leads naturally to the question of what changes could be made to account for the data. Here consideration of alternatives is limited to changes which have been suggested before in other contexts and to changes which recognize that the model in Fig. 3 might still be valid for some results in central vision. Some changes leave the key feature of a single motion mechanism intact. In studies of apparent motion with stimuli defined by different attributes, Cavanagh et al. (1989) noted that some attribute pairings produced weaker sensations of motion than other pairings. They attributed the differences in effectiveness of the pairings either (i) to variations in the visibility of stimuli defined by different attributes or (ii) to the preference of the motion system for or against particular pairings of attributes. In the present context, if A (T-LG ) was made equal to A (T-FG) by setting d’ = 3 for both gratings, it might still be possible that direction discrimination performance for the flicker grating was inferior because T-FG was weighted less than T-LG by the motion mechanism. In fact for the conditions tested, the weight assigned to the T-FG input must have been zero, or close to it because the d’ for direction discrimination of the flicker grating remained near zero for all step sizes (Fig. 4). It is not clear why the weight assigned to T-FG would be near zero in peripheral vision, but on a par with that of T-LG for central vision (Pantle & McCarthy, 1988). Another difficulty with the model in Fig. 3 may be that the assumption that A (T-LG ) is made equal to A (TFG ) when d’ is set to 3 for both luminance and flicker gratings is false. Detectabilities of the luminance and flicker gratings may depend upon signals in form pathways that are separate from the motion pathway (e.g. Livingstone & Hubel, 1987) and not by the amplitude of input signals immediately preceding a motion analysis stage. Under this hypothesis either form signal, T-LG or T-FG, might be completely independent of the preprocessed representations which serve as immediate inputs for a motion mechanism. A less extreme version of this hypothesis would divide each of the pathways of the model in Fig. 3 into parallel channels. Signals from a subset of the channels in each pathway would determine detectability, whereas signals from a different subset would serve as input to the motion mechanism. Again, it is not clear why separate signals might determine detectability and direction discrimination performance in peripheral, but not central vision. It is probably a mistake, however, to define the problem only along the dimension of eccentricity of viewing because Werkhoven, Sperling and Chubb (1993) have shown that

et al.

motion between competing texture patches is not predicted from their detectability even in central vision. A more comprehensive change to the model of Fig. 3 is the addition of a second motion mechanism. T-LG would serve as the input to one motion mechanism; T-FG, the input to a second motion mechanism in a separate pathway (e.g. Braddick, 1974; Petersik, 1989; Cavanagh et al., 1989; Wilson et al., 1992). In a separate pathways model the relationship between T-LG and a later stage motion computation need not necessarily be the same as that between T-FG and its respective motion computation. Moreover, in order to explain the coherence/transparency of moving plaid patterns, Wilson et al. (1992) added a higher-order stage in which the motion signals from first- and second-order motion pathways are linearly summed. The patterns of coherence-transparency which they observed in central and peripheral vision led them to postulate that the motion summing mechanism weighted the second-order motion signal relatively less in peripheral than in central vision. In the present context, this means that, if equally detectable T-LG and T-FG signals resulted in equal direction discrimination performance in central vision (because they led to equally strong signals in the motion summing mechanism), the T-FG signal would necessarily lead to a weaker output from the motion summing mechanism in peripheral vision (because the motion signal in the second-order pathway is weighted relatively less). The pattern of results obtained with luminance and flicker gratings (Pantle, 1992; Expt 2) conforms to Wilson et ul.‘s hypothesis.

CONCLUSIONS

The scientific importance of the present motion results stems from the inability to reduce the differences in the perception of motion of peripherally viewed luminance and flicker gratings to differences in their detectability. The motion results require the pursuit of models and/or assumptions which are different from those outlined in Fig. 3. Whether or not the observed pattern of motion results constitutes a general rule for peripheral vision is not yet known, but regardless, the findings should be revealing of underlying mechanisms. In contrast to the motion findings, the effects of variables like spatial frequency and number of cycles of a grating on the perception of the structure of luminance and flicker gratings appear to be similar provided that the two types of stimuli are made equally visible (contrast equivalent).

REFERENCES Adelson, E. & Bergen, J. (1985). Spatiotemporal energy models for the perception of motion. Journal of the Optical Society of America A. 2, 284-299. Ad&on, E. &

Movshon, J. (1982). Phenomenal coherence of moving visual patterns. Nafure, 300, 523-525. Albright, T. (1992). Form-cue invariant motion proCCSSing in primate visual cortex. Science, 255, 1141-I 143. Bergen, J. & Landy, M. (1991). Computational modeling of visual texture segregation. In Landy, M. & Movshon,J. (Eds),

FLICKER GRATINGS IN PERIPHERAL Computational models of visual processing (pp. 253-271). Cambridge, Mass.: MIT Press. Braddick, 0. (1974). A short range process in apparent motion. Vision Research, 14, 519-521.

Cavanagh, P. & Mather, G. (1989). Motion: The long and short of it. Spatial Vision, 4, 103-129.

Cavanagh, P., Arguin, M. & von Grunau, M. (1989). Interattribute apparent motion. Vision Research, 29, 1197-1204. Chubb, C. & Sperling, G. (1988). Drift-balanced random stimuli: A general basis for studying non-Fourier motion perception. Journal of the Optical Society of America A, 5, 19862007.

Derrington, A. & Henning, G. (1993). Detecting and discriminating the direction of motion of luminance and colour gratings. Vision Research, 33, 799-8 11. Heeger, D. (1992). Normalization of cell responses in cat striate cortex. Visual Neuroscience, 9, 181-197. Lindsey, D. & Teller, D. (1990). Motion at isoluminance: Detection/discrimination ratios for moving isoluminant gratings. Vision Research, 30, 1751-1761.

Livingstone, M. & Hubel, D. (1987). Psychophysical evidence for separate channels for the perception of form, color, movement, and depth. Journal of Neuroscience, 7, 34163468. Pantle, A. (1989). Motion adaptation and peripheral motion discrimination with higher order stimuli. Investigative Ophthalmology and Visual Science (SuppI.), 30, 73. Pantle, A. (1992). Immobility of some second-order stimuli in human peripheral vision. Journal of the Optical Society of America A, 9, 863-867.

Pantle, A. & McCarthy, J. (1988). Two-stage visual motion phenomena. Investigative Ophthalomology and Visual Science (Suppl.), 29,

VISION

173

Prazdny, K. (1986). An asymmetry in apparent motion of kinetic objects. Bulletin of the Psychonomic Society, 25, 251-252. Robson, J. & Graham, N. (1981). Probability summation and regional variation in contrast sensitivity across the visual field. Vision Research, 21, 409-418. van Santen, J. & Sperling, G. (1984). Temporal covariance model of human motion perception. Journal of Optical Society of America A, I, 451-473.

Sperling, G. (1976). Movement perception in computer-driven visual displays. Behavior Research Methods and Instrumentation, 8, 144151.

Stoner, G. & Albright, T. (1992). Motion coherency rules are form-cue invariant. Vision Research, 32, 465475. Turano, K. & Pantle, A. (1989). On the mechanism that encodes the movement of contrast variations: Velocity discrimination. Vision Research, 29, 207-22 1.

Victor, J. & Conte, M. (1992). Evoked potential and psychophysical analysis of Fourier and non-Fourier motion mechanisms. Visual Neuroscience, 9, 105-123.

Waugh, S. & Levi, D. (1993a). Visibility, luminance and vernier acuity. Vision Research, 33, 527-538.

Waugh, S. & Levi, D. (1993b). Visibility, timing and vernier acuity. Vision Research, 33, 505-526.

Werkhoven, P., Sperling, G. & Chubb, C. (1993). The dimensionality of texture-defined motion: A single channel theory. Vision Research, 33, 463485.

Wilson, H., Ferrera, V. & Yo, C. (1992). A psychophysically motivated model for two-dimensional motion perception. Visual Neuroscience, 9. 79-97.

252.

Petersik, J. (1989). The two-process distinction in apparent motion. Psychological Bulletin, 106, 107-121.

Petersik, J., Hicks, K. & Pantle, A. (1978). Apparent movement of successively generated subjective figures. Perception, 7, 371-383.

Acknowledgement-The

first two experiments were part of a thesis submitted by James McCarthy to Miami University in partial fulfillment of the requirements for a Master’s degree.