Face recognition sample library and retrieval method

Abstract

The invention relates to a face recognition sample library and a retrieval method. The face recognition sample library comprises M layers of databases from bottom to top, wherein the Mth layer in theM layers of databases only comprises one sub-sample library node, each of the rest layers comprises a plurality of sub-sample library nodes, the sub-sample library nodes of each layer in the M layersof databases are correspondingly connected with a plurality of sub-sample library nodes of the next layer respectively, one sub-sample library node of each layer is only connected with one sub-samplelibrary node of the upper layer, and the sample data of the sub-sample library node of each layer is from the sub-sample library node connected with the sub-sample library node of the lower layer. Compared with the prior art, the method has the advantages that the face recognition sample library correspondingly connected with the sub-sample nodes of each layer is constructed through the hierarchical model, the scale of the sample library can be effectively increased, mutual independence of various customer groups under a single sample library is avoided, and the false recognition rate and thefalse rejection rate of an existing face recognition algorithm are guaranteed on the basis of a layer-by-layer retrieval mechanism.

Description

technical field [0001] The invention relates to the technical field of data processing, in particular to a face recognition sample database and a retrieval method. Background technique [0002] At present, face recognition technology is mainly divided into two categories, namely 1:1 face verification and 1:N face recognition. Face verification is to compare the similarity by extracting the features of two faces, which is used to judge whether the two input faces belong to the same person. It is suitable for application queries such as identity recognition and similar face query. The current development of face verification mode It is mature and has been widely used in user verification links in various industries. [0003] Face recognition is to find one or more faces with the highest similarity to the face to be retrieved in a large-scale face database. It is necessary to create the facial feature index of the person to be identified in advance, and then search and match ...

AI Technical Summary

Problems solved by technology

Although the sample library can reduce the order of magnitude through sharding technology to reduce the impact of the increase in algorithm time and space complexity due to the increase in the number of samples, the final result of face retrieval is Top1 (the most similar face) in the N comparisons. , this type of method cannot avoid the decline in recognition performance caused by the increase in the size of the sample library to the bottleneck of the algorithm itself

Once the algorithm bottleneck is reached, no matter how the sharding mode and scale are adjusted, the false recognition rate and true rejection rate cannot be guaranteed

Images