A learned distance function for medical image similarity retrieval
Dave Tahmoush · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2009
One fundamental problem remains in the area of image analysis and retrieval: how to measure perceptual similarity between two images. Most researchers employ a Minkowski-type metric, which does not reliably find similarities in objects that are obviously alike. This paper develops a similarity function that is learned in order to capture the perception of similarity. The technique first extracts high-level landmarks in the images to determine a local contextual similarity, but these are unordered and unregistered. Second, the point sets of the two images are fed into the learned similarity function to determine the overall similarity. This technique avoids arbitrary spatial constraints and is robust in the presence of noise, outliers, and imaging artifacts.