Adaptive Support Weight-Based Stereo Correspondence Algorithm for Face Images

摘要

The stereo correspondences of human faces are often very difficult to achieve because of uniform texture, slow changes in depth, and occlusion. In this paper, we introduce an adaptive weight-based stereo correspondence method for face images. We estimate the support weights of the pixels in a given support window based on a color similarity and proximity to reduce the fattening effect. Further, the reference image is segmented using mean-shift segmentation method, and then, self-adaptability measure is used to estimate the initial disparity using correlation-based SSD method. Dissimilarity at a given pixel is then computed using the initial disparity and support weights of both support windows. Finally, the correspondence selected by the winner takes all the method. The experiments are carried out on stereo images of face database and Middlebury database. The experimental results show that the proposed algorithm produces a smooth disparity map while preserving sharp depth discontinuities accurately.