Abstract: Aiming at the problems that sparse matching points can not meet the needs of 3D reconstruction and the traditional dense matching algorithm is unable to match the light dark transform image, this paper proposes a dense matching scheme combining Mahalanobis distance and gradient descriptor. The scheme uses the initial reliable homonymous points to establish a homonymous triangle network, and uses the corresponding midpoint of each triangle as the encryption matching primitive, take descriptor and Mahalanobis distance as two influencing factors to set up score calculation formula, take the person who exceeds the threshold as the matching point, traverse all triangles, update the triangulation network, and repeat the above steps until no new matching point is generated. With the help of network public data set, the experimental results show that the dense matching scheme proposed in this paper can better solve the problem of poor adaptability of traditional algorithms to light and dark transform images, and has good adaptability and stability to a variety of transform images.

Key words: dense matching, Delaunay triangulation, Mahalanobis distance, gradient descriptor, score calculation