An improved active contour model driven by region-scalable and local Gaussian-distribution fitting energy

Images with low contrast, overlapped noise and intensity inhomogeneity of multiple objects make many existing level set methods disabled for image segmentation. To address the problem, an improved active contour model is proposed, driving by region-scalable and local Gaussian-distribution fitting energy for image segmentation. Firstly, we classify regions with similar intensity by utilizing the means and variances of local image intensities. Secondly, we define a new edge stopping functional to robustly capture the boundaries of multiple objects. Finally, we utilize LoG energy term to catch edge information and smooth the homogeneous regions, which can be optimized by an energy function. Experiments results on real and synthetic images validate that our method is faster, robuster and higher accuracy than other major region-based methods for images with multiple objects.