Stereo Matching Algorithm Using a Weighted Average of Costs Aggregated by Various Window Sizes

A window-based stereo matching, which matches pixel values within a window between two images, produces a dense disparity map, and as a result, constructs a dense depth structure. Many algorithms of the window-based stereo matching have been proposed. The conventional algorithms, however, face a trade-off between accuracies of the disparity map in disparity continuity and discontinuity regions due to the window size dependence. In this paper, to solve the issue, we proposed a new algorithm of the window-based stereo matching. In the algorithm, the disparity map is computed using a weighted average of costs aggregated by various window sizes from large to small. Therefore, our algorithm improves accuracy of the disparity map in both disparity continuity and discontinuity regions. In order to evaluate the performance, we have designed C++ programs. The simulation result shows that our algorithm is effective compared to conventional algorithms.