Video object segmentation using an improved level set

In this paper, we propose an improved level set method to solve the tough problem of video object segmentation. In the initialization of each frame, motion estimation compensation (MEC) is employed to evaluate the state of the object in the current frame based on the segmented result of the previous frame. To solve the error-prone at strong edge in the background, background contrast removal (BCR) method is proposed to eliminate the contrast but preserve the edge belonging to the object and background. Our method not only can totally avoid re-initialization in the process of evolution and manually initialization between successive frames but also can produce high quality segmented results in real-time. Experimental results demonstrate the effectiveness and robustness of the proposed approach on the challenging human body segmentation with the complex background.