局部自适应Chan-Vese图像分割模型

Local Self-Adaptive Chan-Vese Image Segmentation Model

经典的Chan-Vese(CV)模型不包含图像的边缘信息,当图像的目标或背景较为复杂时,分割效果并不理想.针对该问题,本文通过结合图像的局部信息对CV模型进行改进,并运用K均值聚类方法计算图像目标和背景区域的灰度均值.其次,在本原对偶理论(primal dual scheme)框架下,本文提出了模型的一个等价形式,并使用半隐式梯度下降法快速求解.实验结果表明,本文模型对合成图像和自然图像都有较好的图像分割效果.

The classic Chan-Vese（CV）model does not include the information of edges.So it only gives some unsuitable segmentation results when backgrounds and foregrounds have complex structures.In order to overcome these faults,we improved the classic CV model by employing the local information of image and computing mean values of backgrounds and foregrounds by the K-means method.Following the framework of the primal dual scheme,we gave the equivalent form of the proposed model and then used semi-implicit gradient method to solve it.Experiments on synthetic and natural images illustrate that the proposed mode is more effective for various kinds of images with complex features.