Locality sensitive hash image retrieval parameter optimization method based on empirical fitting

Abstract

The invention relates to a locality sensitive hash image retrieval parameter optimization method based on empirical fitting, which comprises the following steps of: S1, defining a locality sensitive hash function family H; S2, assuming that k is the number of locality sensitive hash functions and L is the number of hash index tables, when values of L, r and w are determined, calculating a value ofk; S3, taking k functions from the H, and defining a k-dimensional locality sensitive ash function family G; and S4, taking L hash functions from the G, and establishing L hash index tables. According to the invention, a locality sensitive hash image retrieval parameter optimization empirical formula is obtained by a regression analysis method, and by using the empirical formula, calculation steps can be effectively reduced, complexity of parameter optimization of an algorithm can be reduced, and operation efficiency of the algorithm can be improved. Meanwhile, the locality sensitive hash image retrieval parameter optimization method disclosed by the invention is approximate to the theoretical optimum and can enable the algorithm to obtain a high F1 so as to obtain excellent algorithm performance.

Description

technical field [0001] The invention relates to a parameter optimization method, in particular to a local sensitive hash image retrieval parameter optimization method based on experience fitting, which belongs to the field of image processing. Background technique [0002] With the advent of the data age, the processing volume of multimedia data such as images, videos, and audios on the Internet has increased dramatically. The feature dimensions to be extracted from image, video and other data reach hundreds of dimensions or even thousands of dimensions, and these high-dimensional data often show unstructured characteristics. When dealing with high-dimensional data, traditional data processing methods cannot meet the requirements. Algorithms such as data retrieval and semantic analysis pose enormous difficulties. The content-based image retrieval method does not rely on keywords to search, but performs image matching by extracting content features of images. Among them, th...

AI Technical Summary

Problems solved by technology

Coupled with the randomness of parameter selection of locality-sensitive hash function, in the case of a single locality-sensitive hash function, dissimilar data points may be mapped to the same hash value, resulting in errors

Images