In this work, we propose a disparity refinement method to be applied in stereo matching algorithms. It consists of a segmentation process, statistical analysis of grouping areas and a support weighted function to find unknown disparities. We investigate the behavior of this method by comparing it with other post-processing techniques, as the left to right consistency-check. By comparing some of the most common refinement techniques, the experimental results show that or method achieved the lowest erros in non-weighted functions. Furthermore, through a qualitative evaluation, it is possible to note that our method reaches significant results, close to the ground truth maps.