A discontinuity detector based on the pervasive noise in surface property data

A discontinuity detector based on the pervasive noise process near object boundaries produced by certain early vision algorithms is described. Contrary to the assumptions embodied in most discontinuity detectors, the noise in surface property data, such as stereo disparity, is not spatially white, Gaussian noise. This noise is briefly illustrated. To facilitate discontinuity detection, the locations of intensity edges are used as prior estimates to the true discontinuity locations. The discontinuity detector is formulated as a Markov random field; although variations of the approach described should be amenable to hardware implementation