This paper presents an automatic foreground segmentation algorithm for stereo image pair captured by a dual-camera system. Being different from the monocular image, binocular images contain the disparity map between the stereo image pair. For the disparity map is computationally expensive, our approach adopts the residual image with spatial displacement ( , ) to segment the initial trimap automatically. From the residual image, rough region of foreground is clustered as the initial trimap of GrabCut algorithm. Compared with a rectangular region, the calculated trimap is more accurate. After running GrabCut algorithm, the images are segmented into foreground and background layers that comprises of the front objects and back environment. Experimental segmentation results with the original images captured by the dual-camera system indicate that our approach is efficient and promising.