水平集方法(Level Set Method)由于能够将平面曲线的演化过程转变为高一维空间中曲面的演化过程，从而能够有效地解决曲线的拓扑结构变化问题，因此成为图像分割领域重要的研究方向。 针对图像信息的不同，水平集模型主要分为两类，即基于边缘的模型和基于区域的模型。本质上，这两类模型分别利用了图像中的梯度信息和区域信息。第一类模型常见有基于距离正则化的水平集模型—DRLSE模型，第二类模型主要有基于区域特征的水平集模型—CV模型以及基于区域扩展拟合的水平集模型—RSF模型，分别实现了基于这些模型的图像分割算法。通过具体的实验结果，分析了这些算法的优缺点。 提出了基于局部聚类的变分水平集模型—LCVLS 模型，该模型主要是利用了图像中的局部区域信息。该模型首先在图像每个点的局部区域内，采用K-mean 聚类算法对局部区域内的强度值进行聚类，根据K-mean 聚类准则定义了一个关于图像强度的局部聚类准则，再从整体上考虑，定义一个全局聚类准则并使之最小化，从而使整幅图像的分割效果达到最佳。然后将全局聚类准则作为水平集能量泛函的一个能量项，加上正则化项、曲线长度项和惩罚函数项，获得水平集能量泛函。根据公式推导，得到水平集偏微分方程和偏微分离散方程，最后实现了基于该水平集模型的图像分割算法，并通过具体的实验结果证明，该模型对强度不均匀图像分割有着很理想的分割效果。 以医学图像为实验对象，通过基于CV 模型、RSF 模型和LCVLS 模型的图像分割结果的对比，计算出各模型的面积误差率、最小迭代次数和最少收敛时间，从面积误差率可以证明LCVLS 模型比CV 模型和RSF 模型的分割效果更好，而通过最小迭代次数和最少收敛时间可以证明LCVLS 模型比CV 模型和RSF 模型的分割效率更高。最后从总体上来看，通过具体的实验结果证明了LCVLS 模型对医学图像的分割是较为有效的。

Due to The level set method can transform the evolution of plane curve into the evolution of the surface in a higher dimensional space, which can effectively solve the problem about the topology change of curve, therefore it become the important research direction in the field of image segmentation. According to the different of image information, the level set model mainly divided into two categories, the model based on the edge and the model based on region. In essence, these two kinds of models respectively use the gradient information and regional information of the image.The first kind of model includes the level set model based on distance regularized-DRLSE model,The second kind of model includes the level set model based on regional characteristics, and the level set model based on the fitting of the extended area, respectively achieve the image segmentation algorithm based on these models, and through the specific experiment results, we compare and analyze the advantages and disadvantages of each model. This paper proposes a variational level set model based on local clustering. The model mainly use the local information of the image. The model first uses the K-mean clustering algorithm for intensity clustering in the local region of each image point, and defines a local clustering criterion function for the image intensities, then consider the whole image, define a global clustering criteria and make it achieve the minimum to make the whole image segmentation to achieve the best. Then the global clustering criterion as an energy items of the level set energy function, plus the regularization item, curve length and the penalty function, become the level set energy function. According to the formula, we can get the partial differential equation and discrete partial differential equation, and finally achieve the image segmentation algorithm based on this level set model. The specific experimental results prove that this model can solve the problem about the segmentation of images with intensity.