we use a matching process M that is 1 a feature at pixel L in the left eye matches a feature at pixel r in the right eye, and it is 0 otherwise. Within the Bayes approach we define the probability of generating a pair of inputs, W L and W R, given the matching process M by (1)where the second term pays a penalty for unmatched points 2 ,with 3 being a positive parameter to be estimated. C1 is a normalization constant. This distribution favors lower correlation between the input pair of images.