Representation of medical images by visual response functions

A short introduction to the representation of images by visual response functions is presented. The aim is to show how some specific characteristics or features of an image can be found by a machine observer. Edges, flat regions, regions with saddle points, corners, etc. can be found by appropriate convolution of the image with the visual response functions (VRFs). The scale is selected such that noise effects are minimized and/or the desired features are emphasized, followed by a selection operation, usually nonlinear, such as thresholding. In the particular case of medical images, the zero crossings of the Gaussian Laplacian and of the umbilicity operators are being investigated as descriptors of prior information for a specific case.<>