Occlusion refinement for stereo video using optical flow

In various areas of processing video in 3D format the precise occlusion mask is highly required, for example for high quality stereo-to-multiview conversion with background restoration or in quality metrics for stereo video. Stereo occlusions detected by many of existing algorithms have problems with width, alignment with objects borders and smoothness. In this paper we propose a novel technique for occlusion mask refinement using Optical Flow. Constrains on occlusion width, position and smoothness are directly incorporated in our method. Experiments and tests on real video sequences demonstrate satisfactory results. This proves the effectiveness of the proposed technique.