Human perception inspired occlusion detection for stereo vision

Matching cost curves contribute more informative cues than disparity maps to accurate occlusion detection. Inspired by the fact that the human perceived depth of occluded pixels increases with increasing horizontal separation from an occluding edge, this paper proposes a matching cost curve based occlusion detection scheme. An asymmetric occlusion detection method is designed based on the fact that the matching cost curves of the adaptive support-weight approach and human perception have the similar characteristic in occluded regions. Then the matching cost curves based method is combined with the warping constraint to reduce the false negative rate of occluded pixels. Finally, motivated by the nature of occlusion maps, morphology is applied to decrease the false positive rate of non-occluded pixels. Experimental results show that proposed scheme achieves high accuracy.