Optimized block based disparity estimation in stereo systems using a maximum-flow approach

A novel disparity estimation method is presented that increases the robustness of the estimator by replacing the classical approach of dynamic programming with finding the maximum flow in a graph. Once solved, the minimum cut associated to the maximum flow yields a disparity surface for the whole image at once. The results show improved depth estimations as well as better handling of depth discontinuities. Although the running time for solving the maximum flow problem is higher than dynamic programming, experiments have shown that the special topology of the graph, the position of the source and sink and the capacity structure of the edges tend to make the problem easier to solve. However, the main drawback of this approach is the large amount of memory resources required by the classical implementations of the maximum flow algorithm, which make impossible in practice to apply this approach even for the small stereo images with small disparity resolution. Taking advantage of the special topology of the graph and the position of the source and sink we propose an efficient data structure for drastically reducing the amount of memory resources used.