APPLICATION OF SURFACE CONSTRAINT FOR MODELING THE STEREOPSIS OF UNTEXTURED STEREOGRAMS ((János Geier)) Eötvös Loránd University, Budapest, Hungary.

Purpose. In last few years some researchers have discovered untextured stereograms whose half occluded areas and vertical disparities generate subjective depths and contours (Anderson, 1994, Nature). Because untextured stereograms do not contain local texture elements (primitives), so the conventional approach (i.e. matching primitives between the left and right images) is not applicable for modeling this phenomenon. I present a computational model and a computer algorithm for solving the stereo depth problem for both textured and untextured stereograms. This is a modified version of the model I presented last year (ARVO95 #1322). Method. The goal is a direct search for the 3D objects that are in the best correspondence with the left and right images. This correspondence is tested and measured by the ‘surface constraint’. (Given 3D objects serve as the correct solution only if the projection of left and right image to the surfaces of objects is definitely overlapped on the regions that are visible from both cameras (eyes).) The method for searching for 3D objects is a random type method, similar to ‘simulated annealing’. The computer algorithm can make visible the projection of images to the 3D surface. Results. Input stereograms for computer algorithm are similar to ones you can find in ‘Anderson and Julesz (1995) A Theoretical Analysis..., Psych. Rev.’. After correcting the remaining errors by human manipulations, surface constraint indicates that computed measure of overlapping in 3D surfaces is 100%. In this case half occluded regions fall always in the farthest regions of surfaces. Conclusions. ‘Surface constraint’ is a useable constraint for testing the solution of the stereo depth problem for untextured stereograms. The applied searching method requires further development.

Supported by the US-Hungarian Science and Technology Joint Fund, J.F. No. 360. None