Fast stereo matching with super-pixels using one-way check and score filter
Masayuki Miyama · 2017
This paper proposes a novel fast method of depth image generation with super-pixels (SPs) by stereo matching. The proposed method adopts an existing method in which block matching is performed at the center of the SP, and the obtained disparity is applied to all pixels of the SP. Then, in contrast to usual cross check which requires disparity estimations in both left and right directions, we newly introduce one-way check that judges errors using only unidirectional disparity estimation to reduce the amount of calculation. Furthermore, to improve disparities, a score filter on SP basis is newly introduced. The filter multiplies the matching score of the surrounding SP by the weight consisted to the reliability and the similarity, and sums up the weighted scores in the neighborhood. Even if the block size to be matched is reduced, the score filter can suppress accuracy degradation, so that the amount of calculation can be further reduced. As a result of the experiment, the execution time of the proposed method was about 1/160 of full search, about 1/3 of the conventional center matching, and the accuracy deterioration was slight.