Object-based stereo matching using adjustable-cross for depth estimation
Li-Hung Wang, Kai-Lung Tsai, Chung-Bin Wu · 2015
In this paper, an algorithm of object-based stereo matching using adjustable-cross is proposed for depth estimation. The algorithm can generate the depth map of the target objects in the image. It is composed of three steps: Pre-processing, Matching Cost Aggregation and Refinement. In Pre-processing, the Edge Detection is used to remove redundant background information and retain shapes of objects. To improve the shapes of objects, the Object Extension is further applied. Then, the holes generated by previous processing are removed by the Hollow Filling. In Matching Cost Aggregation step, the proposed adaptive window whose size is computed by extracted region of object is used for the Stereo Matching. Moreover, the disparity map is computed and transformed to depth map by the Stereo Matching. Finally, the generated depth map is refined by median filter in the Refinement. The experimental results show that the proposed algorithm can reduce the computational complexity about 91%.