Stereo Image Coding Using Object-Based Global Constraints

We propose a system to perform disparity estimation for a pair of stereo images using object-based global constraints. EEective global constraints are crucial to accurate disparity estimation under the presence of partially occluded areas, illumination diierences, and object deformation. In our method, object contours are rst extracted and their disparities are estimated using a robust tree matching method. The disparities on the object contour are then used as boundary conditions for disparity surface model tting to obtain the overall disparities for the image pair. The employment of disparity at object contour as global constraints not only allows disparity discontinuities along object contours and, thus, improves the sharpness of the reconstructed image, but also make the constraints more accurately reeect the characteristics of the disparity map. In addition, the smoothness of resulting disparity map within an object makes high compression ratio with high reconstructed image quality achievable.