Active Contour Model Based on Local Intensity Fitting Energy for Image Segmentation and Bias Estimation

Abstract

Intensity nonuniformity is one of the common issues in image segmentation, which is caused by technical limitations or external interference. In this paper, a novel region-based active contour model is presented for interleaved segmentation of images with intensity nonuniformity and correction of the bias field. First, we define the local region-based fitting image by using the information of bias field and the intensity, and simultaneously introducing the local difference between the input image and estimated image. Next, a likelihood fitting image energy functional is built in a local region around each point. Then, a level set method is used to present a total energy functional, which contains the level set distance regularization term and the length regularization term. Extensive experiments are conducted on synthetic images and real medical images to demonstrate the advantages of our model over the state-of-the-art methods. Segmentation results show robustness to initialization and noise, as well as significant improvements in both accuracy and execution time.