An improved two-view convergent-style-based disparity estimation algorithm

Disparity estimation is a key problem in stereo video compression. In this paper, an improved two-view convergent-style-based disparity estimation algorithm was proposed. The features of disparity vector in two-view parallel and convergent camera systems were analyzed based on their convergent styles respectively, and how to find the best matching block for the parallel system by a two-step searching method and for the convergent system by a three-step searching method were introduced. The proposed algorithm was tested and compared with the original TZSearch algorithm in JMVC in different types of sequences. Experimental results showed that the proposed algorithm could largely reduce the disparity estimation time with a slight increase in bit rate, while maintaining good image quality when compared with the TZSearch method. Meanwhile, since there was no epipolar rectification in the proposed algorithm, the disadvantages caused by epipolar rectification were avoided properly.