Matching local image feature descriptors

摘要

A computer-implemented method of matching features identified in first (114) and second (115) images captured from respective camera viewpoints (112, 113) related by an epipolar geometry, each identified feature being described by a local descriptor 204. The method comprises: determining view constraints for one or both of the camera viewpoints; using the epipolar geometry and the determined view constraints to define a geometrically-constrained region (118) in the second image corresponding to a first feature (109) in the first image represented by a first local descriptor 204; comparing 206 the first local descriptor with local descriptors of features in the second image, thereby determining 207 respective measures of similarity between the first feature in the first image and the respective features in the second image; identifying, from the measures of similarity between the first feature in the first image and the respective features in the second image, a best match feature to the first feature. The descriptors could be vectors representing pixel characteristics and similarity between vectors could be determined by subtraction.