Edge detection by associative mapping

A new set of template matching edge operators is developed using an associative mapping between ideal step edges defined on a 3 × 3 neighborhood and the orthonormal basis of the neighborhood regarded as a nine-dimensional vector space. Detection of an edge is based only on the confidence in the goodness of fit to a template and the performance does not deteriorate for low amplitudes. The method can be adapted to the specific needs of the user; the set of masks can be reduced trading orientation resolution for amount of computation; and edges can be thresholded adaptively to the local background level. When applied in a dual way, the edge detection procedure provides an estimate of the standard deviation of the noise present in the image. Optimality for step edge detection of one of the 3 × 3 Laplacian operators is shown. The edge images obtained from the new edge operators are suitable inputs into relaxation algorithms based on local consistency.