Active contour model based on partition entropy and local fitting energy

In order to overcome the problem of initialization sensitivity in segmenting images with inhomogeneous intensity, an active contour model based on global and local fitting information is proposed. According to the theory of entropy can measure the degree of heterogeneity of the segmentation regions, the global fitting energy based on partition entropy is established. The introduction of variance makes the global information of the image more fully described. Also, the energy functional is constructed by combining the established global partition entropy energy and the local binary fitting energy. In the energy functional, the contribution of global information and local information is controlled by a dynamic weight parameter, which can improve the convergence speed. At last, the segmentation experiments are carried out on synthetic images, medical images and natural scene images with intensity inhomogeneity. The experimental results show that the proposed algorithm has good performance in terms of sensitivity, accuracy and efficiency.