摘要
为解决基于随机映射的高维向量快速检索方法位置敏感哈希存在的随机性强和内存消耗大两个问题,在E2LSH(Exact Euclidean Locality Sensitive Hashing)的基础上提出了基于多表频繁项投票和桶映射链的快速检索方法。该方法用检索结果构造基准索引矩阵,并对基准索引矩阵进行频繁项投票和校正得出最终索引来降低检索的随机性;桶映射链利用E2LSH的数据划分特性减少检索时读入内存的数据点的数目,以此来降低内存消耗。实验证明该方法能减弱检索的随机性,并有效地降低检索的内存消耗。这对于提高大规模信息检索尤其是图像检索的可行性有着较大的作用。
To solve the problem of strong randomicity and high memory cost of fast retrieval method Locality Sensitive Hashing （LSH） based on random projection, a fast retrieval method is presented based on multi table frequent items voting and bucket map chain on the basis of Exact Euclidean Locality Sensitive Hashing （E2LSH）. The method constructs an index matrix with retrieval vectors, and performs frequent items voting and calibration on this matrix to decrease the randomocity. It also reduces the number of points loaded into memory by making use of the data partition property of E2LSH to decrease the memory cost. The experiments show that this method can decrease the randomicity and efficiently reduce the memory cost of retrieval. This is very important for increasing the feasibility of large scale information retrieval especially image retrieval.
出处
《电子与信息学报》
EI
CSCD
北大核心
2012年第11期2574-2581,共8页
Journal of Electronics & Information Technology
基金
国家自然科学基金(60872142)资助课题
关键词
信息检索
位置敏感哈希
随机性
内存消耗
频繁项投票
桶映射链
Information retrieval
Locality Sensitive Hashing （LSH）
Randomicity
Memory cost
Frequent items voting
Bucket map chain
