Context dependent processing in spatial hyperacuity

OWN989 S5 53.00 + 0.00 Copyrrght i: 1985 Pergamon Press Ltd

ViFion Rt+x.Vol. 25. No. 12. pp. 1959-1992. 1985 Pnnted m Great Britain. All rights reserved

RESEARCH NOTE CONTEXT

DEPENDENT PROCESSING HYPERACUITY PETER MEER

and YEHOSHUA Y. ZEEVI

Department of Electricai Engineering. Technion-Israel (Receked 26 March

IN SPATIAL

institute of Technology, Haifa 32000. Israel

1985; in recked form 24 .Jufi 1985)

Abstract-The importance ofcomparison processes in spatial hyperacuity tasks was investigated by means of a new relative nonalignment discrimination task. Vernier-type configuations nonaligned to either left or right, were presented in succession, and the subject had to decide which configuation exhibited a larger nonalignment. The relative nonalignment thresholds were found to be more strongly dependent on the spatial arrangement of the two lines than the thresholds corresponding to “ordinary” spatial hyperacuity performances. This context dependency is attributed to the active role played in the processing by

comparison mechanisms. Hyperacuity

Local/global

Context dependency

The optical and neurophysiological components, as well as system organization, impose limitations on performance of the visual system. The coarseness of the retinal mosaic, matched to the optics of the eye, is the main factor limiting spatial resolution. (It permits resolution of the order of 30 set arc.) However, detection of relative position in spatial hyperacuity tasks presents thresholds which are an order of magnitude lower than resolution thresholds. This fact implies that hyperacuity tasks must involve additional processing of signals impinging on the retina. Any stimulus employed in hyperacuity tasks can always be divided into a reference and a test component. The subject executes the task by discriminating a relative-measure defined as the difference in one of the common spatial features of the two components. That is, to execute a hyperacuity task, a comparison process is always necessary. The reference and test components can be presented in the visual field simultaneously, as is the case in differential hyperacuity tasks (Westheimer and McKee, 1979; Hadani er a/., 1980; Nakayama, 1981). Alternatively, the stimulus can be defined in different temporal frames, as for example in some spatial interval sensitivity tasks (Westheimer and McKee, 1977; Westheimer, 1981). The precise role played in the processing of a hyperacuity task by the comparison of reference and test components is not yet clear. The available experimentai data do not make it possibie to establish beyond any doubt the strategy employed by the visual system. In one possible strategy, each of the two components is being independently processed with the best hyperacuity precision the system can achieve; the comparison process is only necessary in

Image understanding

order to permit a read-out of the relative-measure. In another strategy, the visual system does not employ the best accuracy for representation of each of the separate stimulus components. The excellent performance in spatial hyperacuity tasks is obtained according to this strategy by a comparison process. In order to investigate the influence of the comparison process on the spatial hyperacuity threshold, we have embedded the same relative nonalignment tusk in stimuli with different spatial structures. From the obtained results we conclude that an active role is required for the comparison process between the reference and test stimulus components. Five, vertically oriented, vernier-type basic configurations were employed (line length 11 min arc). The configurations differed in the relative position of the upper line. This position was described by two parameters: s, the vertical extent of the gap: and I, the horizontal extent of the initial lateral offset (see inset B in Fig. 1). The parameter values characterizing the basic configurations are given in Table 1. and the configurations drawn along the abscissa of Fig. 1. These and other similar configurations are often employed in spatial hyperacuity tasks such as vernier acuity and spatial interval sensitivity. From the thresholds obtained under these circumstances one can deduce a measure of the accuracy achieved by the

1989

Table

1.

Spatial parameters of configurations (in min arc)

Configuration

(I

b

Gap size(s)

0

4

Initial lateral

0

0

-:* 4

the

basic

d

e

0 4

4 4

offset (r) ‘The lower endpoint of the upper line is located below the upper endpoint of the lower line.

I990

Research Sots

v-isual system in the representations of the lateral offsets. Specification of gap sizes and initial lateral offsets of about 4 min arc assures the best hyperacuity performances when using these configurations (Westheimer, i 98 I ). In the relative nonalignment task, both the reference and the test stimulus components have configurations similar to one of the five stimuli described above. The same experimental procedure was repeated for each basic configuration. Ten nonaligned stimuli were derived from a basic configuration. The derived configurations differed only in the lateral locations of the upper line. The new locations were calculated by adding to the initial lateral offset & 1. 22, +3. *5 or 26 units of displacement. The displacement unit was of a few seconds of arc. Thus, two sets of five configurations were generated. The two sets differed only in the direction of nonalignment (left/right) of the upper line relative to the lower one. Stimuli were generated on a Tektronix 608 CRT display equipped with P31 phosphor. Subjects were seated at a distance of 400 cm in front of the screen, with the head immobilized by a forehead and a chin rest. They observed the stimuli binocularly with natural pupils. The lines had a luminance of IO cd/m’, and appeared highly suprathreshold in the dim, diffusely illuminated room. Stimulus duration of 1 set was equally divided into two 500msec intervals. For the first interval a leftward or rightward nonaligned configuration was selected at random. After 500 msec a new configuration belonging to the complementary set characterized by the nonalignment in the opposite side was randomly selected. The subject had to decide whether the leftward or the rightward oriented configuration exhibited a larger nonalignment. This was a dative nonalignment task, assessed through the 2AFC method. The left/right paradigm was preferred over the first/second one in order to minimize the effect of bias toward the last presented configuration. A trial consisted of responses to all 50 possible stimulus combinations which can be derived from a basic configuration. Ten of the stimuli were of equal left/right nonalignment. Responses to these stimuli were used to assess the subject’s bias. The other 40 were employed in reduction of a psychometric curve, specifying the number of correctly discriminated relative nonalignments for 1, 2, 3, 4 or 5 units of displacement, The displacement unit (of a few seconds of arc) was separately specified for each basic configuration, and rougly corresponded to guessing in the subjects’ response. The same value of the unit was used for all three subjects. All three subjects were well experienced in visual experiments. Subject NI had at least 20 trials per basic configuration while subjects A.Y. and P.M. had only IO trials. Data were gathered for several basic configurations in a session, to compensate for inter-

session fluctuations

in subject response. The 7090 correct response potnt was determined from the psychometric curve by probit analysis (Finney. 1971). This value was taken to be the rc+~tire nonnl@nre~rr rfiscriminntion ~hrdroirf for the basic configuration. The standard deviations of the thresholds were less than 15% of the mean. The employed sign-alternating procedure reduces. as much as possible. the similarity betwen the reference and test components of the stimuli. We tried to constrain the visual system to use information explicitly related to the individual components of the stimuli. Note that the extent of the jump of the upper line. considered alone. does not supply any relevant info~ation for the task. Depending on the type of the basic configuration the relative nonalignment threshold employed, exhibited different values (Fig. 1). TO assess the significance of these differences, we measured the “ordinary” nonalignment, i.e. vernier thresholds of the subjects. The experiments were conducted with both vernier basic configurations (a and h), i.e. abutting lines and a 4 min arc gap. The “ordinary” nonaligment thresholds are shown clustered on the left side of Fig. 1 with the open symbols corresponding to configuration (I, and the open symbols marked with a dot corresponding to configuration h. These thresholds have similar values. in accordance with results reported by Westheimer and McKee (1977). However, the relative non~llignment thresholds obtained with the basic configurations a and b differ significantly. Assuming for the moment that the first of the two processing strategies discussed earlier is valid, the visual system would, under these circumstances, have processed the two components of the stimulus independently, resulting in the best possible precision. Ideaf (i.e. noiseless) comparison and decision mechanisms imply that the threshold values (70% correct detection) obtained in the vernier tasks correspond to pure guessing of the relative nonalignment. This is certainly not the case for configuration a, where both the vernier and the relative nonalignment thresholds have the same values of about 8 see of arc. For configuration b, the compound task still resulted in a threshold somewhat superior to that expected based on experiments performed with the basic con~~uration. The best vernier thresholds are close to the best spatial interval sensitivity (Westheimer and McKee, 1977). The similarity is only maintained in the relative nonalignment discrimination task when the interval is defined by vertically overlapping lines (basic configuration c). Unfortunately. we could not employ configurations with larger vertical overlap of the lines because of apparent interline affinity (Ullman, 1979) resulting in perception of a con~guration shift as a whole instead of a jumping upper line. The spatial interval sensitivity threshold does not depend on the length of the lines demarcating the

Research Note

1991

. :

. . .

1, 1 I

B

a

b

c

I

d

I I

I

I

8 I

I

Basic cantigura?ions Fig. 1. Relative nonalignment discrimination task. The subject has to discriminate between two oppositely aligned vernier-type configurations, presented in succession, each for 500 msec (See insets A and B.) The five employed basic configurations are drawn along the abscissa. Solid symbols represent the relative nonalignment thresholds. Open symbols (clustered on the right side of the figure) represent the vernier thresholds for the basic configuration o, and open symbols marked with a dot the vernier thresholds for the basic configuration b. Subject A.Y.: n ; subject NJ.: 0; subject P.M.: A.. The standard deviation is

about 15% of the mean.

interval (Westheimer and McKee, 1977). Nevertheless, the relative nonalignment threshold increased further when the spatial intervals were delimited only by the adjacent end-points of the two lines (configuration d). The observed performance appears to delineate an upper bound. Defining the interval by means of a virtual end-point (basic configuration e), resulted in the same threshold as in the case of basic configuration d. The physical proximity between the lines appears to be a necessary condition for good relative nonalignment performances. Nevertheless, we cannot interpret our data solely through the proximity of the upper and lower lines in the basic configuration. While the adjacent end-points of the two lines are situated at the same distance in configuration b and d, the relative nonalignment thresholds differ significantly. Interpretations based on discrimination of changes in the orientation defined by the two adjacent end-points are also unsatisfactory. In configurations c and e the range of tilts is similar (around the 45 deg meridians), but the relative nonalignment thresholds differ significantly. We conclude that by combining two vernier-type tasks into a relative nonalignment sensitivity experi-

ment, we significantly increased the influence of the spatial structure of the stimulus on the threshold. Although the subject had to discriminate the same spatial feature, the visual system appeared to use, in the relative nonalignment task, different processing strategies for the different basic configurations, suggesting that the processing is context dependenr. By and large, the context dependency runs counter to the insensitivity exhibited by the “ordinary” hyperacuity tasks to configuration structure (Westheimer and McKee, 1977; Westheimer, 1981), or line length (Sullivan er al., 1972; Westheimer and McKee, 1977). Thus the mechanisms subserving the process of comparison between the reference and test stimulus components has to be responsible for the context dependency reported in the relative nonalignment task. The spatial structure of the basic configuration is represented in the visual system only at a global level. Several mechanisms can convey this global information relating the two stimulus components. For configuration a the subject appear to make use of the absolute slope of the configuration (Watt et al., 1983). Internal representation of the difference between the two tilts could be responsible for the best relative nonalignment threshold obtained. Detection

1992

Research Note

of perturbed mirror symmetry (Barlow and Reeves, 1979) could also be considered as a mechanism subserving the relative nonalignment task. Finally. we observe that the lowest threshold values are similar in both the vernier and relative nonalignment tasks. The existence of ;f common bound supports the assumption that hy~racuity performances are ultimately timited by the noise present in the visual system (Mangaubi and Zeevi, 1979; Geisler, 1984; Zeevi and Mangoubi, 1984). ~cknor~ledgewfenu-The research was supported by Technion V.P.R. Fund-Micay Archie Biomedical Research Fund. and by the Samuel and Isabelle Friedman Trust.

REFERENCES

Barlow H. B. (1981) Critical limiting factors in the design of the eye and visual cortex. The Ferrier Lecture 1980. Proc. R. Sot. Land. 8212, 1-34. Barlow H. B. and Reeves B. C. (1979) The versatility and absolute efkiency of detecting mirror symmetry in random dot displays. Vision Res. 19, 783-793. Finney D. J. (1971) Probir ~na~~.~j.~.3rd edn.. Cambridge Univ. Press.

Geisler W. S. (1984) Physical limits of acuity and hyperacuity. J. opt. Sot. Am. d 1. 775-782. Hadani I., Gur M., Meiri A. Z. and Fender D. H. (1980) Hyperacuity in the detection of absolute and differential displacements of random dot patterns. Vision Res. 20,

947-95 1. Mangoubi S. S. and Zeevi Y. Y. (1979) A hyperacuity decision modeI. In Proc. of the IOth IEEE Convention, Tel-Aviv, Vol. 133-2, pp. I-3. Nakayama K. (1981) Differential motion hyperacuity under condition of common image motion. Vision Res. 21, 1475-1482. Sullivan G. D., Oatley K. and Sutherland N. S. (1972) Vernier acuity as affected target length and separation. Percept. Psychophys. 12, 438-144. Ullman S. (I 979) The Interpretation of Visaal Motion. MIT Press, Cambridge, Mass. Watt R. J., Morgan M. J. and Ward R. M. (1983) The use of different cues in vernier acuity. Vlrion Res. 23,991-995. Westheimer G. (19X1) Visual Hyperacuity. In Progress in Sensory Physiology, pp. l-37. Springer, Berlin. Westheimer G. and McKee S. P. (1977) Spatial configurations for visual hyperacuity. Vision Res. 17, 941-947. Westheimer G. and McKee S. P. (1979) What prior uniocular processing is necessary for stereopsis. Inuesr. Opthal. ciwzl Sri. 18, 614-621. Zeevi Y. Y. and Mangoubi S. S. (1984) Vernier acuity with noisy lines: Estimation of relative position uncertainty. Biol. Cvbernet. 50, 371-376.