We present here our approach to the problem of improving the efficiency of stereo depth map computation. The algorithm is applied on rectified images. Graph cut is used for energy minimization. The descriptors used are both SIFT and DAISY. This algorithm produces fast results of approximate disparity maps from two images. The main advantage of our algorithm is its efficiency and reduction of computation time, in spite of an improvement of the error performance. To achieve this, we initially use a sparse global matching technique using SIFT to determine the necessary labels and then find dense correspondence with DAISY.