Belief Propagation Stereo Matching Based on Differential Geometry Constraint of Disparity

The particularities of stereo matching for slant planes and curve surfaces are studied and a differential geometric constraint of disparity is put forward based on image segmentation, The energy function is constructed according to the disparity constraint, which can reflect better the depth structure of visual scene, The classic belief propagation optimization algorithm based on front parallel plane assumption is improved using the disparity constraint the disparities of pixels in segmentation area are optimized iteratively until the energy function is minimized. The experiments on Middlebury Test Set show that the proposed method is more fit for vision scenes of slant planes and even curved surfaces.