Human face quick retrieval method and system

Abstract

The invention discloses a human face quick retrieval method and system. The method comprises the steps of obtaining an eigenvector of an image; inputting the eigenvector to an auto-encoder network; according to the auto-encoder network, performing training and updating to obtain a weight of a full connection layer and a corresponding bias term, and taking the weight and the bias term as network parameters for performing binary hashing on the eigenvector; through the network parameters, establishing a hash index library of the image and obtaining a hash value of the to-be-queried image; and searching out a human face result. According to the method and the system, a deep convolutional neural network is used for human face feature extraction, so that efficient human face feature expression can be obtained. Meanwhile, hash codes are obtained by using the auto-encoder network, and compacter binary expression is obtained based on human face features. In addition, the image similarity is calculated by adopting a Hamming distance of the hash codes, so that the calculation amount is small and the retrieval speed can be increased.

Description

technical field [0001] The present invention relates to the field of deep learning and face image recognition, in particular to a fast face retrieval method and system, mainly based on convolutional neural network and self-encoding network binary hash. Background technique [0002] The face image database in the prior art includes many types, such as, FERET face database, CMU-PIE face database, YALE face database, MIT face database, ORL face database or the like. And the purpose of adopting the face image library is: in the face image library, retrieving similar face images has broad application prospects in face recognition directions such as monitoring and security. [0003] It is known that performing hash coding on the original image can effectively improve the speed of image retrieval. [0004] Such as in the prior art, Chinese patent application number: CN 201310087561.5 a kind of similar face retrieval method based on local sensitive hashing, discloses a kind of face...

AI Technical Summary

Problems solved by technology

[0005] The analysis shows that the shortcomings of this method are: the eyes, mouth and skin color features can not express the characteristics of the whole face well, and the local sensitive hashing method is a data-independent hashing method with strong randomness; To ensure better retrieval accuracy, the required number of coding bits is very long, and the retrieval efficiency is relatively low

[0007] The analysis shows that the disadvantage of this scheme is that the data of training images is less, and the global frequency features are not suitable for face recognition tasks.

Images