基于多Agent进化计算的图像目标识别  

Image Object Recognition Based on Multi-Agent Evolutionary Computation

作　　者：周亦鹏[1] 胡娟[2] 杜军平[3] 

机构地区：[1]北京工商大学计算机与信息工程学院,北京100048 [2]北京政法职业学院信息技术系,北京100024 [3]北京邮电大学计算机学院,北京100876

出　　处：《复杂系统与复杂性科学》2013年第3期55-60,共6页Complex Systems and Complexity Science

基　　金：国家重点基础研究发展计划(973计划)(2012CB821206);国家自然科学基金(91024001;61070142);北京市自然科学基金(4111002)

摘　　要：对图像的多目标识别问题进行研究,提出一种基于Agent进化计算的目标识别方法。该方法通过Agent和Agent群体的协同进化完成对样本图像集的快速搜索和模板匹配。同时,针对目标遮挡的问题,利用遮挡圆来构造模板,实现对遮挡目标的识别。实验结果表明,提出的方法实现了平移不变、旋转不变、尺度不变、镜像不变的识别,而且能够准确地确定出遮挡区域的位置和遮挡区域的大小,同时也大大减小了模板匹配的计算量。Image objects recognition is studied. A new object recognition method based on Agent evolutionary computation is proposed. In this method, the processes of searching object on sample image set and template matching are all achieved by cooperation of Agent groups. Moreover, to solve the problem of occluded object recognition, a new method using occluding circle to construct template is proposed too. The experimental results show that the proposed method has the features of translation invariance, rotation invariance, scale invariance and mirror invariance, at the same time, the occluded area can be located. Furthermore, the computation amount of template matching is greatly reduced.

关 键 词：图像处理 目标识别 AGENT 进化计算 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]