Deformable probability maps: Probabilistic shape and appearance-based object segmentation

We present the Deformable Probability Maps (DPMs) for object segmentation, which are graphical learning models incorporating properties of deformable models into discriminative classification. The DPM configuration is described by probabilistic energy functionals, which incorporate shape and appearance, and determine boundary smoothness, image features consistency, and topology with respect to the image salient edges. Similarly to deformable models, DPMs are dynamic, and their evolution is solved as a MAP inference problem. DPMs offer two major advantages: (i) they extend the Markovian property in the image domain to incorporate local shape constraints, similar to the known internal energy of deformable models, and therefore provide increased robustness in capturing objects with fuzzy boundaries; (ii) during their evolution, DPMs update the region statistics, and therefore they are robust to image feature variations. In our experiments we evaluate the DPMs’ performance in a variety of images, while we compare them with existing deformable models and classification approaches on standard benchmark datasets.