基于NMF图像摘要的Web商品图像检索  

Web Commodities Image Retrieval Based on the Image Abstract with NMF

作　　者：戴飞[1] 袁贞明[1] 余天豪[1] 

机构地区：[1]杭州师范大学信息科学与工程学院,浙江杭州310036

出　　处：《杭州师范大学学报（自然科学版）》2012年第3期264-268,共5页Journal of Hangzhou Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(60773051);教育部211重点工程项目(201003010)

摘　　要：提出了一种基于非负矩阵分解图像摘要的图像检索方法.算法首先进行平滑、边框裁剪等图像正规化处理,然后利用非负矩阵分解其灰度图像得到特征系数矩阵,再将其量化构造图像摘要,最后基于最短图像摘要的汉明距离进行图像相似性检索.实验分析结果表明,该摘要对图像多类修改均表现出良好健壮性,查全率和查准率都较基于颜色直方图和基于SIFT特征的方法有较大提高,可较好地满足Web商品图像检索要求.The paper proposed an image retrieval method based on image Abstract factorization with non-negative matrix.The algorithm dealt with images normally by smoothing and frame cropping,obtained the characteristic coefficient matrix by factorizing grey level images with non-negative matrix,constructed the image Abstract by quantizing the matrix,and retrieved the image similarity based on the minimum Hamming distance of image Abstract.The results show the proposed image Abstract is robust under various image modification.Compared with the methods based on color histogram and SIFT,the recall ratio and precision ratio of the proposed approach is much better.The performance can meet the demands of web commodities image retrieval.

关 键 词：图像检索 图像摘要 非负矩阵分解 电子商务 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]