A variational level set model with kernel metric induced local image fitting energy

Abstract

Active contour based methods are effective models for image segmentation. However, they always suffer from the limited performance due to the presence of noise and intensity inhomogeneity. To solve this problem, a kernel metric induced local image fitting (KLIF) variational model is proposed in this paper. Firstly, a kernel metric induced local fitting image (KLFI) is introduced by minimising a kernel metric based energy. The combination of the kernel metric and the local fitting image enables the model to be more robust to the noise and intensity inhomogeneity. And then, using the KLFI, a variational level set model that is a squared l2 distance between the KLFI and the original image is constructed. Two regularisation terms are employed in the model to keep the level set function to be stable during the evolution. At last, an alternating iterative algorithm combining with fixed-point iteration and gradient descent of three-step time-splitting is introduced to solve the proposed model. The experimental results show the effectiveness of the proposed model for image segmentation in the presence of noise and intensity inhomogeneity, and demonstrate the competitive performance over several state-of-the-art variational models in term of accuracy and robustness.