Image segmentation using a novel dual active contour model

Lingling Fang, Xiyue Liang, Chang Xu, Qian Wang

Abstract: For some complex images, low contrast, intensity inhomogeneity, and blurred edges are common phenomena, which inevitably cause difficulties in image segmentation. As a popular image segmentation method, the active contour model (ACM) is often used to solve the above problems. However, the ACM is highly dependent on the initial evolving curves, which makes the model unstable and complex in the actual image segmentation. In this paper, a novel dual active contour model (DACM) is proposed to segment images, which integrates region and edge information to obtain accurate segmentation. Thereinto, the two contours are initialized and evolved simultaneously. The proposed DACM can use region-based and edge-based information, which can handle images with complex structures. For region-based DACM, uniformity among the object pixels and background difference is interlinked to provide an evolving force. Here, uniformity among the object pixels is constructed based on the color reward strategy. For edge-based DACM, the adjustable weighting parameter is set based on image gradient information of two evolving curves. The edge-based DACM can make the evolving curves move inward or outward adaptively. The proposed method is evaluated on various synthetic and real images and accurate segmentation results are obtained. Besides, the state-of-the-art methods are compared with the proposed DACM and an in-depth study of this novel method is given, which denotes that the proposed model can be applied to different types of complex image segmentation.