A vector fuzzy search method and system based on geometric space division

Abstract

The invention provides a vector fuzzy search method based on geometric space division. The method comprises a vector index storage method and a vector search matching method. The index storage method comprises the steps of: A, converting the similarities between numerical feature vectors into distances between two vectors of a geometric space for measurement; B, selecting a search precision of fuzzy search and converting the precision into a corresponding vector space distance, set as Dm; C, with the Dm obtained in the B process as the unit length, performing space division on all vectors and numbering all obtained space blocks to obtain a serial number set ID (x); D, for a to-be-stored vector a, determining that the serial number of the storage block to which the vector (a) belongs is set as ID (a) according to ID mapping relationships; E, creating a Hash Map with ID (x) as the key value and, during storage, judging whether an ID (a) linked list obtained in mapping has a corresponding item in the Hash Map; F, repeating A-E to complete the storage of all the vectors. The method has the advantage of extremely small calculation quantity.

Description

technical field [0001] The invention belongs to the technical field of data retrieval, and in particular relates to a vector fuzzy search method and system based on geometric space division. Background technique [0002] At present, intelligent analysis and processing based on video and pictures have been more and more widely used in various fields. In these applications, a typical problem is: how to find out whether a particular object is in this data set and how many eligible objects are in this data set by searching in a large-scale set. [0003] In such applications, the object to be retrieved (such as a face image) is usually preprocessed before the search, and the object is described by a numerical vector (integer or floating-point vector), which is usually called For the eigenvalues ​​of the original object, this preprocessing process is called a structuring process; the eigenvalues ​​obtained in this process are usually comparable, that is, the results obtained betw...

AI Technical Summary

Problems solved by technology

In these applications, a typical problem is: how to find out whether a specific object is in this data set and how many eligible objects are in this data set by searching in a large-scale set

[0006] 1. Although the operation has been converted, the mathematical meaning is no longer equivalent, which reduces the accuracy of the search

[0007] 2. Although the number of similarity matches in the search process has been reduced by optimizing the index structure, there is still a problem that the amount of calculation increases with the growth of the scale because the matching requirements have not disappeared.

[0008] 3. The support for fuzzy matching is lost or the accuracy of fuzzy matching is reduced

Images