基于混合特征的近似重复图像检索方法  

A Near-Duplicate Image Retrieval Method Based on FusedFeature

作　　者：孙昌[1] 朱允斌[1] 金城[1] 肖晓滨 

机构地区：[1]复旦大学,上海市智能信息处理重点实验室 [2]上海赛为信息技术有限公司

出　　处：《微型电脑应用》2015年第9期21-24,4,共4页Microcomputer Applications

基　　金：国家科技支撑计划项目(编号:2012BAH59F04);复旦大学卓学计划项目资助

摘　　要：仅基于SIFT特征的近似重复图像检索有较高的错误率,无法满足大规模数据的检索精度要求。此文提出了一种基于混合特征的近似重复图像检索方法,该方法将SIFT特征点附近的边缘特征检测结果与SIFT描述子一起,用汉明嵌入的方式生成比特串。对于比特串的匹配结果,使用了一种评分加权方法,提高了结果精度。在Oxford5K数据集上的实验结果表明,该方法有效地提高了SIFT描述子的抗干扰能力,降低了近似重复图像检索的错误率,减少了在线查询所需时间,具有较高的工程应用价值。Near-duplicate image retrieval based on SIFT feature only has high error rate,which cannot satisfy the precision requirement in large-scale image dataset.This paper proposes a method to extract local edge feature,which is suitable for large-scale near-duplicate image retrieval.This method combines the local edge feature and the SIFT descriptor to generate sketch by Hamming embedding.The sketches are matched in the search step,and the match score is calculated by a weighted method according to the Hamming distance,which can improve the precision of the result.Experiments on the Oxford5 K dataset show that this method improves the robustness of SIFT features and decreases the error rate of large-scale near-duplicate image retrieval.This method also reduces the time cost of online querying,which means it is suitable for real applications.

关 键 词：近似重复图像检索 边缘特征 SIFT特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]