Abstract:

A subpixel matching algorithm for curve correspondences from images is presented. Firstly, the image curve is discretized as the pixel points. Then the initial point correspondences on the correspondence curves between the different images are performed by dynamic programming. At last, the precise point correspondences on the correspondence curves between the different images are implemented by using the conjugate gradient based on the initial matching. Compared to previous methods, the reconstruction precision is improved since the manual matching between different images is not involved by self-automatic precise extracting the curves. Experiments show that the reprojection error is remarkably reduced by the presented algorithm.

Key words: close-range photogrammetry, image matching, subpixel, curve correspondence, reprojection error