A stereo visual pattern image coding system

A binocular stereo system for images coded by Visual Pattern Image Coding (VPIC) is presented and evaluated. The use of VPIC for stereo vision applications is motivated by the capability of its coding primitives to reflect meaningful physical properties of projected real-scene surfaces: high-information edge regions and uniform regions. An algorithm for spatial matching of VPIC primitives is proposed. It involves the solution of a complex optimization problem satisfying VPIC-related constraints. A dense disparity map is obtained and then used, together with one of the VPIC coded images of the stereo pair, to predict the second stereo image. The proposed stereo system allows a very high compression of the overall stereo information. Experimental results are presented to illustrate the performance of both stereo VPIC matching and predicting algorithms in test images. (C) 1999 Elsevier Science B.V. All rights reserved.