Robust objectsegmentation using low resolution stereo

Object segmentation based on the joint use of depth and color data produces superior results than can be achieved with either data source alone. Conventional object segmentation methods using depth data give priority to processing speed. So the shape accuracy is not enough to use object-based video coding, or they have some restrictions for background and camera movement. In this paper, we propose a robust object segmentation method, which can deal with both static foreground object and non-static background. In proposed method, we use depth information to make initial segmented regions and to improve accuracy at following modified active contour model. We also consider using a low resolution camera to set up stereo-pair camera at small cost. Our experiments show that the proposed method extract accurate object shape from low-resolution stereo-pair still images.