Eliminating Errors in Image Matching Caused by the Failure of Heuristics

Applications of Computer Vision in Medical Image Processing

Volume

Papers from the 1994 AAAI Spring Symposium

Contents

An algorithm for determining correspondences between images has been developed and applied to the analysis of 2-D Electrophoretic gels. Various heuristics - both derived from the comparison problem and the specific application - are incorporated into the algorithm and result in a fairly high level of accuracy. Consistency checking mechanisms, based on exploring violations of a basic symmetry relation between correspondences, have been developed and incorporated into the algorithm to achieve higher accuracy. The effect of these consistency checking mechanisms is to make inferences concerning the identification and cause of errors due to the (relatively rare) failures of heuristics used in earlier stages of the algorithm. Experiment analysis showing the benefit gained by this consistency methodology is presented.