Abstract
Gold and colleagues (2000) employed a classification image (CI) technique to show that interpolated regions between inducers influence how observers discriminate? fat? and ?thin? Kanisza squares. Here, we extend the technique to explore contour interpolation in dynamic illusory figures.
In Experiment 1, two subjects (one naive) observed black disks translating left to right on a gray background behind the top and bottom corners of an (otherwise invisible) gray stationary figure. Interpolation was necessarily spatiotemporal in that inducing events at each corner occurred at different times. The observers' task in three conditions was to discriminate i) pairs of rectangles with real contours, ii) Kanisza (illusory) rectangles, or iii) fragmented figures (designed to produce no interpolation). Convolution and quantization techniques (Ahumada & Beard, 1998) showed that a) subjects were sensitive to regions between the inducers in the illusory (interpolation) condition but not in the fragmented control condition; and b) the illusory and real conditions yielded comparable CIs.
In Experiment 2, disks were stationary and fourteen subjects (over 28,000 trials) judged whether a horizontally translating contour was convex or concave. Noise fields were as before (truncated, Gaussian white; power=3.7 * 10-5 deg2) except that there was a new field for each frame of the sequence. Six times more pixels reached significance in the illusory region than predicted by chance alone (p<.001), and a contour can be seen in certain frames. Thus even when an illusory contour moves, and even when its inducing events are sequential, subjects are influenced by pixels along interpolated contours.
Supported by National Eye Institute EY13518 to PJK.