A new real-time algorithm for Multiview Disparity based 3D-object segmentation

With the development of video communication system, more and more attention has been put on the implementation of varieties of functionality associated with content. Thus, the object segmentation algorithms featured both semantic and automatic are tremendous necessary. In this paper a new real-time algorithm for 3D-object segmentation algorithm based on disparity is proposed. Its kernel disparity matching methodology is developed on base of a general pixel-based segmentation algorithm known as Dynamic Programming on Disparity Space Image. With the introduction of weighted feature span, the computational complexity is reduced apparently. The proposed algorithm is tested on both real captured stereo sequence and synthetic multiview sequence, and gets satisfying result in generality, quality and complexity.