Hellinger Distance Based Active Contour Model

Image segmentation is a major field of research in image processing.The active contour model is a process that transforms the evolution of the curve into an update of the level set and seeks to minimize the energy functional with variational principle.Due to its good accuracy,the active contour model is a current research hotspot.Aiming at the problems of image intensity inhomogeneities and the robustness of the noise and the initial contour,this paper proposes two new region-based active contour models.The main work is as follows:A local region intensity fitting(LRIF)model is proposed.Because the local binary fitting(LBF)model and the local image fitting(LIF)model both have the sensitivities about the initial contour and noise.To solve this problem,based on two different local fitted images,the Hellinger distance is used to measure the similarity between the fitted image and the original image in order to classify the local pixels,and then a new variational energy functional is constructed.Experimental results show that the proposed model can improve the segmentation effect to inhomogeneous intensity images,strengthen the robustness of noise and reduce the dependence on the initial contour position.A local hybrid energy fitting(LHEF)model is proposed.Because the local intensity clustering(LIC)model needs to perform multiple convolutions,the segmentation efficiency is reduced.For this problem,we fully utilize the local information of the image.Firstly,based on the extended fitted image and the square fitted image,the LIC model and the LIF model are fused by the Hellinger distance.Secondly,the Hellinger distance between the extended fitted image and the original image and between the square fitted image and the square of the original image are used to classify the pixels.According to the Hellinger distance,divide the closer pixels into one category and the farther ones into one category.And then,combined with the bias field,the model can segment the inhomogeneous intensity images.Finally,the distance regular term and the length fitting term are combined to ensure the smooth of the curve in the evolution process.Experimental results show that the model not only can enhance the segmentation efficiency,but also improve the disadvantages of the LIF model about the sensitivity of the initial contour.