量化误差的索引检索方法

Index Retrieval Based on Quantization Errors

为了提高乘积量化的近邻检索方法的检索精度,本文提出了量化误差的索引检索方法。该方法以子空间量化误差为依据进行子空间码字的选择,以获取高质量的乘积量化候选码字,提高基于乘积量化的近邻检索方法的检索精度。最后,通过把索引检索方法应用于5种基于乘积量化的索引结构上,在3个基准数据集上进行实验。实验结果表明,索引检索方法可以提高检索精度的有效性。

In order to improve the retrieval accuracy of the product quantization nearest neighbor retrieval method,this paper proposes an index retrieval method based on quantization errors(IRQE).The method selects subspace codewords based on subspace quantization errors to obtain high-quality PQ candidate codewords and improves the retrieval accuracy of PQ-based NNS methods.Finally,on the three benchmark data sets,combined with five PQ-based NNS methods,the retrieval accuracy has been significantly improved compared with the original methods,which shows the effectiveness of the IRQE method.