双倍比特量化近似查询索引算法研究

Research on Approximate Query Index Algorithms with Double Bit Quantization

本文提出双倍比特量化与非对称距离的近似查询索引。首先,设计了一种双倍比特量化方法,通过把特征的每一维数据量化为两个比特二进制码,增加特征之间的区分性。然后,研究了非对称距离算法,通过计算浮点型查询特征与特征库中二进制码的距离,对海明空间下的最近邻进行重排序,以提高索引的查询精度。基准数据集上的实验表明,双倍比特量化与非对称距离的方法使最近邻查询精度提高15%~25%。

In this paper,we proposed an approximate query index based on double bit quantization and asymmetric distance.First of all,a double bit quantization method was designed to increase the distinction between features by quantizing each one-dimensional data into two bit binary codes.Then,the asymmetric distance algorithm was studied.By calculating the distance between the floating-point query feature and the binary code in the feature library,the nearest neighbor in Hamming space was reordered to improve the query accuracy of the index.Experiments on the benchmark data set show that the accuracy of nearest neighbor query is improved by 15%~25%by using the method of double bit quantization and asymmetric distance.