An active contour model based on hybrid energy and fisher criterion for image segmentation

Abstract

Image segmentation (Lin et al. 2018, Shoulin et al. 2018) is the basic link of image processing and machine vision, so it is very important to study this content. Because the image can be interfered by many factors in the process of imaging, the image is usually with some problems such as heterogeneous, edge blur, noisy, which makes low accuracy of traditional segmentation methods. In recent years, the active contour model (Ben Rabeh et al. 2017) has achieved great success in image segmentation, which is widely concerned by researchers. The active contour model can be divided into two categories: parameter active contour model (Hanbay and Talu 2018) and geometric active contour model (Khamechian and Saadatmand-Tarzjan 2018). The geometric active contour model can be subdivided into two categories: dynamic contour model based on edge (Amini et al. 2004) and active contour model based on region (Soudani and Zagrouba 2017, Elisee et al. 2017). For the edge-based active contour model, the results are not ideal when it segments weak edge and the images with noise, the robustness of the initial contour lines is also poor. To this end, the scholars propose an active contour model based on region. There are some representative region-based active contour models, such as chan-vese model (CV) and Piecewise Smooth model (PS). The model is used to guide the curve evolution by using the information of pixel grey variance in the curve. And it does not depend on the gradient information of the image, so it can deal with the weak edge of image better. Shyu et al. (2012) introduced the Retinex theory into the active contour model, which was commonly used for image segmentation. The segmentation procedure was then guided by the image intensity and light reflection. In order to solve the proposed model efficiently, this work developed a new fast Split Bregman algorithm. The results were better. Xiaomin and Tingting (2017) presented an alternative criterion derived from the least squares projection twin support vector machine (LSPTSVM) for image segmentation. The proposed model treated image segmentation as pattern classification problem, and hence tried to seek the projected axis and centre for the foreground and background intensities, respectively. With level set representation, the discriminative function of LSTSVM was incorporated into the energy function of the active contour model, and derived the contour evolution accordingly. Experiment results demonstrated that this model held higher segmentation accuracy and more noise robustness. Le and Savvides (2016) proposed a novel joint formulation