An Adaptive Window Stereo Matching Based on Gradient
Yan Ping He, Pei Wang, Jie Fu · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2013
The key of local matching algorithm is the selection of similarity measure function and window's size, this paper through the two aspects to do improvement. Due to the traditional adaptive window algorithm is relatively complex, this paper extracted a relatively simple adaptive window algorithm, that is to say the window size based on the gradient points, and adopt a new measure function instead of the SAD function, not only to reduce noise effects but also to improve matching accuracy. In addition, in order to get a better disparity map, we adopt a regional seed propagation method which based on smooth constraint and a hypothesis that adjacent pixels with similar color should have the same disparity. Experiments show that the algorithm is compared with the traditional matching algorithm of adaptive window, computational complexity is reduced and the matching accuracy improved. proposed an adaptive weighting method, which can effectively structure match cost, greatly reduce the matching ambiguity, the disparity map can compete with global optimization result, but the computation is big, this method could not reflect the efficient advantages of local algorithm. Federico carefully analysis of the deficiency of the adaptive weighting method (8), using color segmentation information improves the weighting function, improve the accuracy of the algorithm, but further increased the operation cost. Therefore, under the condition of the trade-offs between efficiency and precision, this paper proposes adaptive windows of all different shapes and sizes based on different gradient. Compared with traditional methods, its computational complexity is reduced, but the matching precision is improved. In order to get a better performance, we use the method of seed propagation method, and to introduce a new similar measure function which has good noise resistance. The experiments show that the algorithm can improve the quality of image matching and compared with the traditional algorithm, computational complexity is reduced.