Relaxational refinement of intensity ridges

A frequent goal in the early processing of images is the location of step discontinuities of intensity. However, in certain image domains the important features are present not as steps but as narrow intensity extrema, or ridges. As with step-edge detection, the raw ridge information can be characterised using specialised filters. Subsequent postprocessing must then be performed to extract consistent connected ridge contours. The authors describe a novel and powerful evidence combining approach to the postprocessing operation. This approach is based on a Bayesian probabilistic relaxation framework that has already been used to design a successful tool for step-edge refinement. Its application to ridge refinement requires some significant model additions including the development of a probabilistic description of ridge filter response. The resulting ridge detector is evaluated for the test application of locating road networks in aerial infra-red data