Abstract
Accurate stereo matching is a difficult problem because of differences of the appearance and/or the shading of target objects between two views. To avoid this difficulty, we have proposed a two-view method based on shading and stereo information without explicit stereo matching. This method takes a high computational cost, since it needs iterative updates of many unknown variables including not only the depth but also various shading parameters. In this study, by removing a specular reflection component from the observed image as a preprocessing, we simplify the above method, and hence we aim to lower the computational cost with improving the accuracy of recovery.
