Stereo Matching Algorithm Based on Improved Census Transform and Fusion Cost

To solve the problem that traditional Census transform relies too much on window center information is easily disturbed by noise, and has poor matching effect in weak texture and occlusion area, a stereo matching algorithm based on improved Census transform and fusion cost is proposed. In the cost calculation stage, the improved Census transform is integrated with the sum of absolute difference and gradient feature to calculate the initial cost. In the cost aggregation stage, a guided filter is introduced to constrain the regular coefficients with edge perception weights. The winner-takes-all algorithm is used to calculate the parallax value, and the parallax is optimized by left-right consistency detection, uniqueness constraint and median filtering to get the final parallax map. The proposed algorithm is tested by using Middlebury platform standard images. The experimental results show that the average mismatching rate is 4.91%, and the matching accuracy is improved compared with the traditional Census algorithm.