The Correspondence Problem: An Uncertainty Reasoning Approach

摘要

In this paper, the Correspondence Problem in stereo vision is approached under the Dempster-Shafer Theory for uncertainty reasoning. New results are presented, as a continuation of previous work, using different image conditions to evaluate the robustness of the method. The evidences of correspondence are the contextual and structural features of the pair of points to be corresponded. The Dempster's rule of combination makes the fusion of the evidences in searching for the point that maximizes the Belief on the combined evidences. The results show the robustness of the model in establishing the correspondence in situations for which illumination and/or focus changes occur from one image to the other.