Orientation: Rc Rs, Lc Ls, Rc Ls, Lc Rs . Spatial Frequency: Lc Ls, Hc Hs, Hc Ls, Lc Hs .
Stimuli and Apparati. • Viewsonic Profession PS775 17-inch monitor connected to Dell Dimensions XPS R450 • Custom graphics software • Spatial Resolution: 2 pixels/min 2 • Mean luminance: 19.8 cd/m • Stimuli were 40-min patches of 4 cpd gratings with annular, in-phase 4 cpd grating surrounds (20 minute width). • Four stimuli were intermingled in each session. Both center (c) and surround (s) contained a discrimination cue (right or left of vertical for orientation judgments; lower or higher frequency for spatial frequency judgments)
Observers. • 3 females, 2 males • 19 –22 years of age • Emmetropic or corrected by lenses • All observers were naïve as to the purpose of the experiment
Methods
Fine spatial discriminations made on center-surround stimuli show lateral masking effects. Unlike overlaid (spatially coincident) masking patterns, these effects are asymmetric; the surround affects the center but not vice versa. Also unlike effects found with overlaid patterns, there seem to be no higher level circuits that sum (or difference) responses from cues in two different components, demonstrated in a two -cue, single-response configural (Olzak & Thomas, 1991) effect test. In the current study, we tested the hypothesis that cues to discrimination in center and surround portions of a stimulus are in fact processed independently, with no excitatory or inhibitory interactions, and no correlation in noise. We used a concurrent response paradigm and associated multivariate SDT analyses to identify the underlying response space and to test whether an independence model adequately describes how two cues to discrimination, one in the center and one in the surround, are processed. The stimuli were patches of 4 cpd vertical sinusoidal grating, arranged in a center-surround configuration (40 min center, 20 min width surround). Observers either made discriminations based on spatial frequency or on orientation, in different experiments. A cue to discrimination was presented in both center and surround on each trial. Four stimulus types were created and intermingled in a single session of 160 trials (40 of each stimulus): 1) both center and surround tilted left slightly (or were of slightly lower frequency), 2) both tilted right, 3) center titled left, surround titled right, and 4) center tilted right, surround tilted left. Observers made separate decisions on center and surround patches following each trial, rating their certainty that each component was tilted left or right on a 6-point rating scale. The results supported the notion of no higher-level summing circuits, but strongly rejected a bivariate-Gaussian independence model.
Table number 2 RL 45 2 0 3 5 3 0 45 0 1 5 6 0 0 23 5 3 0 106 93 95 16 0 0 219 151 96 0 5 0 111 193 128 0 0 4 Table number 3 LR 6 1 0 94 136 58 0 13 0 133 137 121 0 0 16 167 130 81 3 8 12 56 0 0 4 7 6 0 93 0 1 3 1 0 0 73 Table number 4 RR 6 0 0 0 1 2 0 30 0 2 2 1 0 0 50 6 2 2 0 1 4 222 0 2 2 5 2 0 514 1 0 5 3 0 3 493
Original frequency tables Table number 1 LL 649 1 0 0 0 1 0 443 0 0 3 1 0 0 138 7 0 1 6 9 13 42 0 0 4 4 4 0 22 0 4 2 3 0 0 3
Observer ALD
Orientation
Data for Fig. 1
Procedures • Starting with a very easy orientation discrimination, track performance as the orientation differences decreases • Blocks of 160 trials repeated over days until each stimulus had been judged 6001200 times, for a total of 2400-4800 trials per observer • On any trial, one stimulus appeared for 500 ms, followed two 5000 ms (max) response periods separated by a short tone that sounded immediately after 1st response was made. • Calculate daily d’, average over the 5 replications
Task • Make two discriminations on each trial, responding sequentially to the center and surround using a signal detection rating scale (1-6) certainty as to whether stimulus A or B had been presented • Order of responding was counterbalanced across observers
Methods cont’d
2
Figure 1
Right
Another indication that the graph is nonsense is that according to this picture, discrimination would be better between LR and RL (negative diagonal) than between LL and RR (positive diagonal). In a single-response 2AFC discrimination paradigm, ALD’s performance was significantly better with the LL vs. RR discrimination.
Center Judgment
Left
The frequency with which each combination of responses was tallied in four 6x6 matrices. We attempted to fit a bivariate Gaussian model to each observer’s data. While our algorithm returned some pictures that almost looked reasonable, it did not converge for some observ2 ers. For those that did converge, Χ goodness of fit values were in the millions, indicating that the data were most definitely NOT well fit by a bivariate Gaussian.
Results
Miami University of Ohio 2 University of California Berkeley
Surround Judgment
Abstract & Introduction
1
1
Lynn A. Olzak Patrick J. Hibbeler and Thomas D. Wickens
1
Results cont’d
This research was supported in part by NIH grant EY13953 to LAO. Many thanks Amanda Dues, Charity Boos, Brock O’Mara, Chad Bechtle and Chrissa Morris.
Acknowledgment
Olzak, L. A. and Wickens, T. D. (1997) Discrimination of complex patterns: orientation information is integrated across spatial scale; spatial-frequency and contrast information are not. Perception, 26, 1101-1120.
Reference
These results differ markedly from those obtained with overlaid second cues. In the latter case, a bivariate Gaussian model (with or with or without additive or inhibitory interactions or correlated noise fit very well and provided insight into the nature of interactions between components (Olzak & Wickens, 1997). The center-surround case is not so simple. We are working to develop a more appropriate model for these data.
Discussion
If observers are wrong in classifying which type of pair was shown, they tend to use the strategy used with the opposite type (positive diagonal or opposing (upper left and lower right quadrants.
What decision rule are observers using? In some observers’ data, a similar pattern occurred. It seems that at least some observers are using a two-part decision strategy. They first classify the stimuli as same sign (LL,RR) or different sign (LH,HL). If they are correct, same sign-decisions lie close to or on the diagonal, and observers tend to get the “right” answer (upper left for the first table), and different-sign decisions lie in the upper right or lower left quadrants. Observers are less certain of the different-sign decisions, as they do not all lie on the negative diagonal.
Processing Cues to Discrimination in Center-Surround Stimuli
Right Left