Optimising edge detection
W.J.C. Witts, Georg Otto · 2003
A feature detector is described that bases its research on a set of probability distributions that describe the situations in which the feature can be found. The result is a measure of the probability of the feature occurring at a particular point. The probabilities required to operate this detector can be automatically derived from test images, although it can be altered to detect any kind of feature. The output from this detector is a set of probabilities that represent the likelihood of and edge occurring at each point in the image. These need only to be summed to find the probability of an edge over an interval, and, in a similar fashion, images can be subsampled in order to find probable edges to subpixel accuracy.>