Fast Disparity Range Estimation and Its Applications

Disparity range estimation is very important in stereo matching. An appropriate disparity range can increase the precision and speed of stereo matching. This paper proposes a fast disparity range estimation method based on matching-cost search and image subdivision. It evenly divides input images into several sub-blocks, and searches using matching cost to find out which sub-block is having the maximum/minimum disparity. After that, the maximum/minimum sub-block is recursively divided into smaller sub-blocks, until all current sub-blocks have the same disparity. To deal with image blocks with week textures, detecting their disparity reliabilities is applied via matching-cost diagram. Experimental results show that our method can achieve 27.7% reduction rate of search space while preserving 93.7% hit rate on average. Compared to traditional methods, it can get a more accurate disparity range. Moreover, the gained disparity range reduces the running time of stereo matching by 20%-45% while decreasing the average false-match rate.