A Shape-Based Approach to Robust Image Segmentation

Abstract

We propose a novel segmentation approach for introducing shape priors in the geometric active contour framework. Following the work of Leventon, we propose to revisit the use of linear principal component analysis (PCA) to introduce prior knowledge about shapes in a more robust manner. Our contribution in this paper is twofold. First, we demonstrate that building a space of familiar shapes by applying PCA on binary images (instead of signed distance functions) enables one to constrain the contour evolution in a way that is more faithful to the elements of a training set. Secondly, we present a novel region-based segmentation framework, able to separate regions of different intensities in an image. Shape knowledge and image information are encoded into two energy functionals entirely described in terms of shapes. This consistent description allows for the simultaneous encoding of multiple types of shapes and leads to promising segmentation results. In particular, our shape-driven segmentation technique offers a convincing level of robustness with respect to noise, clutter, partial occlusions, and blurring.