Occlusion and visible background and foreground areas in stereo: a Bayesian approach

Efficient techniques are introduced in this paper for the identification of the occlusion and visible background and foreground areas in a noisy stereoscopic image pair. Three different Bayes decision methods are tested for this purpose. The first, and uses three hypotheses for the formulation of the Bayes decision rules, adopting the right image as a reference. After performing a dual-Bayes decision test having each time as a different image of the stereo pair as reference, consistency checking is added to these tests to form the second method. Finally, four compound hypotheses are used in the third method, which is the most accurate but also the more detailed and computationally involved of three. Experimental results illustrating the performance of the techniques are presented and evaluated