一种改进的图割目标分割算法  

An Improved Algorithm of Graph Cut Segmentation

作　　者：汤依婷 韩彦芳[1] 

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《电子科技》2016年第10期43-46,共4页Electronic Science and Technology

摘　　要：为了减少图像目标在分割过程中受到噪声、复杂背景等因素的影响,将图像的多特征信息引入到图割算法中,提出了一种结合图像的多特征信息图割目标分割方法。该方法先选取像素点的多种图像特征组成特征向量,并对已做好标记的目标和背景种子点的特征向量分别进行FCM聚类,然后分别计算各像素点与这两类种子点的各聚类中心的最短欧式距离,并据此信息完成对能量函数的构造,最终运用最大流/最小割的方法得到图像分割的结果。其与传统图割算法相比,分割结果有了明显改善。实验结果表明,该算法具有有效性。A combination of multiple image feature information and graph cut algorithm is proposed for segmenting target by introducing multiple feature image information into the graph cut algorithm to reduce the negative influence on the target image of noise, complex background and other factors in the segmentation process. First, multiple image features are selected to compose the feature vectors, and the feature vectors of labeled target and background seed points are clustered by FCM. Second, the shortest distance of each pixel to the cluster center of each seed point of the two types is calculated, according to which the energy function is constructed. Finally, the maximum flow min- imum cut method is used to get the results of image segmentation. Experimental results show that the proposed algorithm significantly improves the segmentation results over the traditional graph cut algorithm.

关 键 词：图割 图像分割 特征向量 FCM聚类 最大流最小割 

分 类 号：P391.41[天文地球—地球物理学]