Face image retrieval method and device based on deep learning and Hash coding

Abstract

The invention relates to a face image retrieval method and device based on deep learning and Hash coding. For the problems of insufficient facial feature expression and insufficient feature discrimination in existing facial image retrieval, a facial space network and facial space loss are provided to automatically mine a facial region with discrimination in a facial image, so that the influence ofbackground information is reduced. Meanwhile, the Hash network learns the internal relation between the face features and the Hash codes, the face images are mapped into the Hash codes, and the calculation and storage cost of retrieval is remarkably reduced. The multi-scale face feature channel enhancement module in the Hash network enhances the dimension of strong distinguishing power in the face features. According to the alternate training strategy provided by the invention, the two networks are organically fused together, the information loss between face feature extraction and hash codegeneration is reduced, the distinguishing power of the face features is enhanced, meanwhile, the distinguishing power of the generated hash codes is enhanced, and the accuracy of face retrieval is improved.

Description

technical field [0001] The invention belongs to the fields of information technology and image retrieval technology, and in particular relates to a face image retrieval method and device based on deep learning and hash coding. Background technique [0002] Face image retrieval refers to, given a face image, finding an image consistent with the identity of the given face image in a huge database of face images. This work has important practical application value in many fields such as mobile payment and security. Face image retrieval at this stage can be roughly divided into traditional retrieval methods and retrieval methods based on deep learning. [0003] Traditional face image retrieval is generally a two-stage process: the first stage extracts the features of the face image; the second stage further processes the extracted features to improve the retrieval effect. For example, the Chinese patent (application number: 201110430327.9, publication number: CN102567483B) use...

AI Technical Summary

Problems solved by technology

However, existing methods usually treat face features of each dimension equally

This will make it difficult for the learned hash code to express the small differences between faces, which will affect the retrieval accuracy

Images