Automated detection of cowhide defects using Markov random field techniques

We deal with the automated detection of cowhide defects. First we introduce a local contrast measure in order to take into account efficiently salient features for discrimination between defective and no defective regions. An Ising model serves as prior model for reinforcing region connectedness. We then propose a two-stages optimization algorithm: first a very fast one-pass step producing a rough but correct estimation; second a Markovian deterministic relaxation step providing the final much more accurate estimation