A boundary localisation algorithm consistent with human visual perception

This paper describes an algorithm, based on psychophysical models of edge localisation, for computing an edge location in presence of blur. The algorithm successfully finds the most likely location at which a human observer would place the edge. For simulated image data the maximum displacement between the boundary placing by the human subjects and by the algorithm was 1.7 pixels. The application of the algorithm to real medical images also showed good agreement, with average radial displacement of 3.3 pixels for skin lesions and 2.7 pixels for mammographic lesions. As the algorithm is based on general principles believed to underlie human visual perception it should be generally applicable.