We present a new adaptive stereo matching method based on Rank transform. When Rank transform is used to measure local intensity, which can tolerate a significant number of outliers, adaptive window selection and support-weight methods based on edge detection are employed to approximate disparity, so an accurate result can be acquired. The experimental results are evaluated on the Middlebury dataset, demonstrating our method outperforms other methods on standard stereo benchmarks.
展开▼
