一种基于精确欧氏位置敏感哈希的目标检索方法

针对目标检索问题，常用方案是视觉词典法（bag of visual words，BoVW），但传统的BoVW方法具有时间效率低、内存消耗大以及视觉单词同义性和歧义性的问题. 针对这些问题，该文提出一种基于精确欧氏位置敏感哈希（exact Euclidean locality sensitive Hashing，E2LSH）的目标检索方法. 首先，采用E2LSH 对训练图像库 的局部特征点进行聚类，生成1 组支持动态扩充的随机化视觉词典组；然后，基于这组词典构建视觉词汇直方图和索引文件，并由tf-idf 算法对词频向量重新分配权重；最后，将目标直方图特征与索引文件进行相似性匹配，完成目标检索. 实验结果表明，相比于传统方法，该方法较大地提高了检索精度，对大规模数据库有较好的适用性.

Object Retrieval Based on Exact Euclidean Locality Sensitive Hashing

The problem of object retrieval is often addressed with the BoVW (bag of visual words) method. There are several problems in the traditional BoVW such as low time efficiency and large memory consumption,and synonymy and polysemy of visual words. In this paper, an object retrieval method based on exact Euclideanlocality sensitive hashing (E2LSH) is proposed. E2LSH is used to hash local features of the training dataset,and a group of scalable random visual vocabularies is constructed. Then, the visual vocabulary histograms and index files are created according to these random vocabularies. The term frequency vectors are weighted with tf-idf strategy. Similarity matching between histogram of the query object and index files is made to accomplish object retrieval. Experimental results show that accuracy of the proposed method is substantially improved compared to the traditional methods. The method is applicable to large scale datasets.