Segmentation of inhomogeneous foreground and background intensity objects using a probability density function based data term and nonparametric shape priors

In this paper we consider segmentation of inhomogeneous foreground and background images using nonparametric shape priors. Non-homogeneity of foreground and background regions of objects to be segmented complicates the process of segmentation. Low quality of images and noise can be considered as reasons of this problem. Furthermore these regions themselves can be textured. One approach proposed for solving challenging segmentation problems involves the use of shape priors. Progress has been made on segmentation of in low quality images, by using shape. However, since most existing shape-based segmentation methods are based on the assumption of homogeneity of background and foreground regions, they do not perform well under the types of conditioned described above. We propose a segmentation approach that learns and uses the probability density functions of the inhomogeneous regions as well as the shapes of the objects to be segmented. Our approach produces successful results on images composed of inhomogeneous regions.