Maximum-likelihood classification of image edges using spatial and spatial-frequency features

It is generally well accepted in the image analysis field of research that the geometrical characteristics of intensity edges are related to the different physical processes that gave rise to them. Therefore, an important task for computer vision is the recognition of the shapes of image edges. However, the many attempts at performing this task on the basis of edge local properties only have so far failed to do so. The author presents a method for classifying different types of intensity edges which uses Gabor elementary functions as local visual filters and the maximum likelihood scheme of classification. Results obtained by the application of this method to a real polyhedral image are presented and discussed.<
>