Graph Cut Based Continuous Stereo Matching Using Locally Shared Labels

摘要: We present an accurate and efficient stereo matching method using locally shared labels, a new labeling scheme that enables spatial propagation in MRF inference graph cuts. They give each pixel region set of candidate disparity which are randomly initialized, spatially propagated, refined for continuous estimation. cast the selection locallydefined labels as fusion-based energy minimization. The joint use cuts has advantages over previous approaches based on fusion moves or belief propagation, it produces submodular deriving subproblem optimality, powerful randomized search, helps to find good smooth, planar maps, reasonable natural scenes, allows parallel computation both unary pairwise costs. Our is evaluated Middlebury benchmark achieves first place sub-pixel accuracy.