Video object segmentation and disparity estimation in stereoscopic coding

Abstract

In this paper, a new method of video object segmentation and disparity estimation is proposed for stereoscopic videoconference. The object in left view is first extracted by motion detection combined with spatial segmentation based on improved watershed algorithm, while the corresponding object in right view is extracted by contour correspondence from the left one. After that, precise disparity estimation is implemented for some significant pixels of the object, while disparities of other pixels of the object are interpolated with the precise ones. Compared with traditional block-based algorithms, more precise and smooth disparity field can be finally acquired by the proposed method.