基于蚁群优化多层图划分的彩色图像分割方法  

Color image segmentation based on multi-level graph partitioning using ant colony optimization

作　　者：葛亮[1] 杨竣铎 

机构地区：[1]重庆大学计算机学院软件理论与技术重庆市重点实验室,重庆400044

出　　处：《计算机应用研究》2015年第4期1265-1268,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61073058;61201347);重庆市科委自然科学基金计划资助项目(cstc2012jj A40011)

摘　　要：为了消除基于谱聚类的归一化切分图像分割中聚类参数对分割结果的约束,提出了一种基于蚁群优化的多层图划分算法来进行归一化切分,进而对彩色自然景观图像进行分割。该算法将代表图像的相似度图作为蚁群的栖息环境,在归一化割准则的指导下,通过蚂蚁的觅食行为将相似的顶点逐渐聚集在一起,从而以多层的方式完成图划分。为了降低图像分割的计算量,利用超像素对图像进行预处理。实验对比表明,该算法消除了归一化切分分割结果对聚类参数的依赖,并提高了归一化切分分割的准确性和速度。In order to eliminate the restraint of clustering number to segmentation result of spectral clustering based normalized cut image segmentation,this paper proposed an ant colony optimization based multi-level graph partition algorithm which was used to segment color natural landscape image. The proposed algorithm regarded similarity graph corresponding to image as ant colony＇s habitat,and then grouped similar vertices into partitions gradually under the guidance of normalized cut criterion using ant＇s foraging behavior,which completed graph partition problem by a multi-level way. To reduce calculated quantity of image segmentation,it utilized an effective super-pixel algorithm to preprocess original image. Contrast experiment shows that the proposed algorithm eliminates the restraint,while improves accuracy and speed of image segmentation based on normalized cut criterion.

关 键 词：彩色图像分割 归一化切分 蚁群优化 多层图划分 超像素 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]