%0 Journal Article
%T 一种利用局部结构信息的加权哈希图像检索算法<br>A Weighted Hashing Algorithm Based on Local Structured Information for Image Retrieval
%A 张胜杰
%A 查宇飞
%A 李运强
%A 李寰宇
%A 杨源
%J 西安交通大学学报
%D 2016
%R 10.7652/xjtuxb201610012
%X 针对图像检索领域中现有哈希算法仅考虑数据的全局信息,同时平等地对待每一维投影数据,导致得到的哈希码不能很好地保持原始数据的相似性的问题,提出利用图像数据局部结构信息的加权哈希(WLSH)算法。该算法同时考虑原始图像空间的局部结构信息和投影数据的各维方差,首先利用数据之间的相似性构建关系矩阵,进而获得原始图像数据的局部结构信息;然后通过迭代量化的方法寻找最优的正交旋转矩阵,使得投影以后的量化误差最小;最后通过权重矩阵平衡各维方差,保证每一维的单位编码信息相同,从而实现对原始数据最优的保距映射。在图像库上的实验结果表明,WLSH算法应用于图像检索时在查准率和查全率上比主成分分析？驳？代量化方法分别提高了3％和2％。<br>A weighted local structured Hashing (WLSH) method is proposed to address the problem in the field of image retrieval that most existing Hashing methods only consider the global information and treat each projected dimension equivalently, which leads to a problem that the binary codes cannot efficiently preserve the data similarity. The proposed method considers local structure information of original image data and the variance of each projected dimension simultaneously. An affinity weight matrix is built to describe the relationship between data points and to acquire local structure information of the original image data. Then, an iterative quantization is used to find an optimal orthogonal transformation matrix and to minimize the quantization error. Finally, a weighted matrix is used to balance the variances and to guarantee equivalent information of each Hashing bits, thus the data similarity is effectively preserve. Experimental results based on some large scale datasets show that the precision and recall of the WLSH algorithm are improved by 3％ and 2％ over principal component analysis？？iterative quantization, respectively
%K 投影函数
%K 局部结构信息
%K 迭代量化
%K 平衡各维方差
%K 图像检索&lt
%K br&gt
%K projection functions
%K local structure information
%K iterative quantization
%K balance the variance
%K image retrieval
%U http://zkxb.xjtu.edu.cn/oa/DArticle.aspx?type=view&id=201610012