引力搜索算法优化脉冲耦合网络的图像检索方法

Image Retrieval Method Using Pulse-Coupled Network Optimized by Gravitational Search Algorithm

启发于脉冲耦合网络(PCN)在视觉特征表示方面的优势,提出使用引力搜索算法(GSA)优化脉冲耦合网络(PCN)来提取图像的视觉特征,对PCN的参数使用优化机制来提高所获取的特征质量,由此来提高基于内容的图像检索(CBIR)的分类和检索结果.首先对学习的图像用PCN生成特征码;然后计算特征码间的距离,距离变量作为适应度函数的输入;最后利用引力搜索算法优化PCN的几个变量,进行参数更新.在Caltech256和Corel数据库上的实验结果表明提出方法的有效性,相比于改进的相关反馈方法(IRF)、颜色边缘结合离散小波变换方法(CE-DWT)和色矩结合局部二进制模式方法(CM-LBP),提出的方法检索精确度至少提高了5%,查全率提高4%左右.

Inspired by the visual features represented advantages in pulse coupling network（PCN）,the method using gravitational search algorithm（GSA）to optimize pulse coupling network（PCN）to extract visual features is proposed,in which the parameters of PCN is applied to improve the quality of the acquired characteristics by optimization mechanism,thereby improving the classification and searching results of content-based image retrieval（CBIR）.Firstly,signature is generated by PCN using learning images.Then,the distance between the signature is calculated,and distance is being as the input of fitness function.Finally,gravitational search algorithm is used to optimize several variables of PCN,updating the parameters.The effectiveness of proposed method is verified by the experimental results on Caltech256 and Corel database,compared with method of improved relevance feedback（IRF）,color edge combined discrete wavelet transform（CE-DWT）and color moments combined with local binary pattern（CM-LBP）,the proposed method improves the retrieval accuracy by 5% at least,and the recall accuracy improves about 4%.