Abstract：For solving the incorrect matching problem in computer vision

a stereo matching algorithm is proposed based on scanline optimization in an energy minimization framework and the best variable window as match cell. Firstly

a dissimilarity measure was introduced which combined intensity and gradient information. Secondly

according to cost function composed of the average measurement error

the variance of the errors and the error bias to larger windows

the best window of each disparity was chose as matching cue

and scanline optimization method was used to search optimum matching paths under the constraint that intensity variation accompanies depth discontinuities. Finally

a dense disparity map could be obtained by tracing the best paths. Experimental results show that the proposed algorithm not only has better performance for processing the large low texture areas and depth discontinuities to get more smooth and accurate disparity maps

but also reduces computation cost which is almost three-fourths less than that of original algorithm.