Abstract
Finding the occluding contours of the objects in real 2D retinal images of natural 3D scenes is done by determining, which contour fragments are relevant, and the order in which they should be connected. This task, until recently, has been considered to be insoluble. Spatially-local operations, which are used by almost everyone, never guarantee producing globally-optimal solutions. Spatially-global operations are also impractical because of combinatorial explosion. We developed a spatially-global model that finds the closed contour represented in the image by solving a shortest path problem (SPP) that uses a log-polar representation of the image; the kind of representation known to exist in area V1 of the primate cortex. This works because the shortest path in a log-polar representation favors smooth, convex and closed contours in the retinal image, with the minimum number of gaps. Furthermore, this approach is practical because finding a globally-optimal solution to SPP is computationally easy. Our model was tested in a psychophysical experiment in which the subject was presented with a fragmented convex or concave polygon target among a large number of unrelated pieces of contour (distracters). The density of the pieces of contour was made uniform all over the screen to minimize spatially-local cues. The orientation of each target contour fragment was randomly perturbed by varying levels of jitter. Subjects were required to draw a closed contour on a screen representing the target. The subjects’ performance was nearly perfect when the jitter-level was low. Their performance deteriorated as jitter-levels were increased. The performance of our SPP model was as good as our subjects’ with the convex polygons. It was not as good with the concave polygons until we revised it by adding a local interpolation at its front-end. Our revised model can detect both convex and concave polygons as well as our subjects do.
Meeting abstract presented at VSS 2015