This thesis work is mainly motivated to solve the stereo correspondence problem. As solving the correspondence problem could result a precise disparity map allowing the interpolation of intermediate views in 3DTV, the disparity map estimation receives nowadays constant research interests by lots researchers worldwide. According to the state-of-art, the colour segmentation-based stereo matching algorithm achieves best performance among all algorithms. This thesis work dedicates to use image's colour segmentation information for the region's disparity modeling by polynomial function, and re ne the nal disparity map via a region merging process. The parameters of local matching algorithm have been studied and we address the at- tention at disparity polynomial model tting, disparity map re nement by region merging process in this thesis.