动态小生境人工鱼群算法的图像分割  

An image segmentation method based on dynamic niche artificial fish-swarm algorithm

作　　者：刘恋[1,2,3] 常冬霞[1,2,3] 邓勇[4] 

机构地区：[1]北京交通大学信息科学研究所,北京100044 [2]北京交通大学计算机与信息技术学院,北京100044 [3]北京交通大学北京现代信息科学与网络技术北京市重点实验室,北京100044 [4]中国科学院软件研究所,北京100190

出　　处：《智能系统学报》2015年第5期669-674,共6页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(61100141);中央高校基本科研业务费资助项目(2013JBM021);中央高校基本科研业务费专项基金资助项目(2012RC044)

摘　　要：为了克服传统基于聚类的图像分割算法需要指定聚类数目以及依赖初始值等缺点,提出了一种基于动态小生境的人工鱼群算法的图象分割方法。该算法将图像分割问题转化为根据图像像素特征对像素的自动聚类问题。采用更为简单的个体描述方式,每条人工鱼表示一个分割区域的一个可行解,并对进化过程中的人工鱼进行动态的划分小生境,每个小生境对应了图像分割问题中一个分割区域。通过对鱼群行为的模拟及种群的动态划分实现了对图像分割问题的分割区域中心和区域数的同时进化,实现了一种新的聚类算法,并实现了对图像的自动分割。实验结果表明:该算法可以自动地估计分割的区域数,并获得较好的分割性能。In order to overcome the defects in the traditional clustering-based image segmentation algorithm, e.g., it needs to specify the number of clusters, it is sensitive to initial value, and so on, an image segmentation method based on dynamic niche artificial fish-swarm algorithm （DNAF） is presented in this paper. In the new algorithm, the image segmentation problem is transformed into an automatic pixel clustering process based on the pixel features of the image. A simpler representation is adopted, each artificial fish represents a single feasible solution of one seg- mented area. Moreover, the dynamic identification of the fish niches is performed at each generation to automatical- ly evolve the optimal number of regions. Each fish niche corresponds to one segmentation region in the image seg- mentation problem. Therefore, the proposed DNAF algorithm implements simultaneous evolution in the center of the segmentation region and the optimal number of regions through simulation on the behaviors of fish swarm and the dynamic division of population. It thereby achieves a new clustering algorithm and automatic segmentation of an im- age. Experiment results demonstrate that the DNAF algorithm is able to automatically estimate the number of the segmented regions, and an excellent segmentation performance can be attained.

关 键 词：人工鱼群算法 图像分割 聚类 动态小生境 进化计算 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]