Volume image registration by template matching

摘要

A template-matching approach to registration of volumetric images is introduced. The process automatically selects a dozen highly detailed and unique templates (cubic or spherical subvolumes) from one image and locates the templates in another image of the same scene. The centroids of the corresponding templates are used as corresponding control points and the best four correspondences that minimize an error criterion are used to determine the translational and rotational parameters needed to register the images. Use of different similarity measures in template matching and determination of coordinates of corresponding control points with subpixel accuracy are also discussed.