Active Contour Model with Entropy-Based Constraint for Image Segmentation

For most existing image segmentation methods based on Active Contours, the contour evolutions tend to get trapped in local optima and thus cause the segmentation methods sensitive to contour initialization and deficient in dealing with noisy and inhomogeneous regions. Aiming at this problem, we propose an Active Contour model with Entropy-based Constraint (ACEC). Specifically, the entropy-based constraint is constructed in the form of Relative Entropy to measure the gap between the intensity distribution of contour-segmented regions and the global prior of foreground and background. This constraint actually offers a global guidance for contour evolution and is helpful to achieve the global optima. Incorporating the global entropy-based constraint together with the region scalable fitting term which draws upon local region information, the energy function of active contour model can be reformulated to utilize both local and global region intensity distributions for image segmentation. The experimental results show that, for the images containing inhomogeneity and blurred object boundaries, the ACEC model achieves better performance comparing with the region-scalable fitting model (RSF) and the active contour model without edges (CV).