Fusion of texture, color and gradient information for stereo matching cost computation

Abstract

Stereo matching is widely used in 3D reconstruction, automatic driving, image focusing and so on. The common local stereo matching algorithm is based on color and gradient features to calculate the matching cost. This paper presents a new matching cost calculation method which fuses texture, color and gradient information to reduce the error rate effectively, and obtain better optimization result. Based on the HA algorithm, we make improvements. This paper also opens up a new perspective of stereo matching, taking into account the information of multi-feature space, and enriches the intrinsic link between the various information of the image. The four standard data sets from the Middlebury website are tested, and HA algorithm results are compared, with the experimental result showing that the proposed method is more accurate.