Horizontal adaptive disparity estimation scheme for stereoscopic images

Stereo image compression is more and more important because of new display technologies and the needs of 3D movies. As a video sequence, a pair of stereo images is very similar to each other. Therefore, there are usually a lot of redundancies between them. To improve the compression efficiency, an effective method to estimate the target image from the reference image is needed. In this paper, based on some peculiar properties of stereoscopic images, a new 3-state fast block searching algorithm for disparity estimation is proposed. It applies the horizontal search with a fat rectangular search window and uses the thresholding and prediction scheme. Moreover, the technique of variable block size mode is also adopted. This novel algorithm can achieve a performance close to (even a little better than) that of full search, while only 2.7% computation cost is required. In addition, simulations show that the entropy for coding disparity vectors can also be greatly reduced compared to other state-of-the-art searching methods.