Implicit Active-Contouring with MRF

In this paper, we present a new image segmentation method based on energy minimization for iteratively evolving an implicit active contour. Methods for active contour evolution is important in many applications ranging from video post-processing to medical imaging, where a single object must be chosen from a multi-object collection containing objects sharing similar characteristics. Level set methods has played a fundamental role in many of these applications. These methods typically involve minimizing functionals over the infinite-dimensional space of curves and can be quite cumbersome to implement. Developments of markov random field (MRF) based algorithms, ICM and graph-cuts, over the last decade has led to fast, robust and simple implementations. Nevertheless, the main drawback of current MRF methods is that it is intended for global segmentation of objects. We propose a new MRF formulation that combines the computational advantages of MRF methods and enforces active contour evolution. Several advantages of the method include ability to segment colored images into an arbitrary number of classes; single parameter which can control region boundary smoothness; fast, easy implementation, which can handle images with widely varying characteristics.