Based On Variational Level Set Image Segmentation Algorithm

Image segmentation plays an important role in pattern recognition and computer vision, where partial differential equation based algorithms for image segmentation has been wildly studied in recent years. This kind of algorithms can mainly be divided into edge-based and region-based models. Edge-based model utilizes the edge information, such as gradient, curvature, etc. to split, to segment the images. However, it cannot deal with the changing of curve topology. Region-based segmentation model generally utilizes the level set to deal with the evolution of the curve in a higher dimension. Therefor, the curve can be divided or merged adaptively. Region based active contour model has a great development during last decades.In this paper, we focus on region based active contour models, and improve the models to get better segmentation results. We firstly introduce some basic conceptions of image segmentation, including some state-of-the-art and some mathematical foundations. Then we focus on level set based image segmentation algorithms. The experimental results show the drawbacks of each method. Finally, we propose two improved algorithms.The work and results achieved as follows:(1) A new variational level set image segmentation algorithm is proposed by introducing the local information into the classical level set algorithm based on the global information. Since the CV model can get the global optimum result and the LBF model introduce the local information with a Gaussian kernel, we combine these two model and propose the CV-LBF model. The proposed algorithm is robust to contour initialization, and can get the global optimum results simultaneously. Finally, we derive our model with multiphase level set.The experiments show the effectiveness of the proposed algorithm.(2) By introducing a edge detection factor into the energy function, a multi-contour with single level set model is proposed. This model can separate different gray scale multi-target with single level set, and detect the edges of each target for the utilization of the edge detection factor. The experimental results show that the proposed model has a remarkable effect to the partition the targets with weak edges.