© Sahim Giray.Stereo matching is an important and popular field of computer vision. Numerous researchers worldwide are devoted to enhancing the effectiveness of stereo matching applications. In stereo matching, determining the costs of matching is a critical step. This step generates a cost volume that quantifies the similarity of pixels, and thereafter, it is processed further to generate the final disparity map. The purpose of this study is to improve stereo matching performance by fusing two different cost-volumes, namely Census and MC-CNN. The Census transform and Hamming distance are one of the most frequently used cost functions in conventional approaches. Besides, a matching cost volume generated using a deep learning technique called MC-CNN has been shown to extract more reliable features from images than conventional approaches. Thus, both of these cost computation strategies have a number of advantages and disadvantages. By including deep learning as a cost-volume, the advantages of these two distinct cost-volumes complement one another, resulting in a better cost-volume prior to applying the smoothing operation (e.g. Semi-global matching or More-global Matching). Our findings indicate that fusing cost-volumes improves the stereo matching performance of nearly all of the benchmark stereo images we tested in the Middlebury dataset.