Video object segmentation using depth and saliency map

We consider the problem of moving object segmentation in the video sequences. This paper presents a new scheme based on depth information and saliency map. The proposed algorithm is a two-step method: First, the depth information captured by Kinect is used to extract the player index data. After getting the initial mask, we design a narrowband pixel classifier to increase the accuracy of mask boundary. Second, we try to distinguish the objects in the same depth level with saliency map model and skeleton information. Experimental results show that the proposed method can achieve satisfying segmentation results for various video sequences with different background.