In this paper, we present a novel approach to incorporate geometric shape priors in region-based active contours, inudorder to provide more robustness to noise and occlusions. We define as shape descriptor a set of Legendre momentsudcomputed from the characteristic function of the shape. Such a representation is invariant with respect to someudgeometric transformations and can handle topologically complex objects. The shape prior is then defined as a functionudof the distance, in terms of descriptors, between the active contour and a reference shape. We derive the evolutionudequation that minimizes the prior energy, using a rigorous mathematical framework. Experimental results show theudability of the geometric shape prior to constrain an evolving curve to resemble a target shape. We finally introduce theudnew shape prior into a two-class segmentation functional and show its benefits on segmentation results, in presence ofudocclusions and clutter.
展开▼
