Stereo matching with pixel classification and reliable disparity propagation

In this paper, we propose a novel high-speed stereo matching algorithm using pixel classification and reliable disparity propagation. While the research on stereo matching has such a long history and many state-of-art strategies have been introduced in recent years, the contradiction between the quality and the time consumes has not yet been solved. Our stereo method tackles this problem with two key contributions. First, we classify all the pixels into two categories: consecutive pixels and isolated pixels. When we perform matching cost aggregation, different supports are constructed for different types of pixels. For a consecutive pixel, an orthogonal local support skeleton is adaptively constructed. For an isolated pixel, we build an adaptive binary window. Second, we simultaneously conduct the matching cost aggregation and reliability detection. Once a reliable disparity is found, we propagate it to the whole support region. Experiments show that this algorithm can significantly reduce the computational complexity and ensure the accuracy of the result at the same time.