Single Sample Face Recognition Method Based On Feature Point Multi-scale LBP And Virtual Samples And Its Application

With the rapid development of science and technology,artificial intelligence can be applied in all aspects of people's work and life.Face recognition(FR)is one of hotspots in artificial intelligence field,and its applications include financial and mobile intelligent terminals.In some real scenarios,it is often possible to obtain a single face image from recognition.In this situation,the performance of traditional face recognition methods will be significantly decreased,and it is urgent to solve the problem of face recognition based on single sample.To solve this problem,this paper conducted the research of single sample FR(SSFR),and the main contribution of this paper is listed as follows:(1)At first,this paper analyzes the principle of face detection,face alignment and correction in face recognition,and introduces the implements of SSFR.Then,based on the shortcomings of traditional FR algorithms,the LBP-based algorithm and PCA-based algorithm are analyzed in details.For the problem of insufficient training samples under single sample conditions,the sample expansion methods such as mirror transformation,symmetric face expansion and sliding window expansion are analyzed in details.(2)In order to extract the discriminative features on training samples with inadequate number of training samples in SSFR,a new method is proposed based on feature points multi-scale LBP and virtual samples.This method first generates virtual samples by mirroring to expand the amount of training samples for each person,and then extracts multi-scale LBP from 49 key feature points of face image to extract high-dimensional effective discriminant features of the local and global extents from face images.Finally,the KNN classifier is employed to classify it.Experiments were carried out on three public face databases.Compared with other state-of-art FR algorithms,the proposed method has a significant superiority in recognition rate.(3)In this paper,a face authentication system based on resident identity card is developed.This system mainly includes: an online information collecting module of ID card chip,a face image collecting module of ID card,and face authentication module based video stream.In this paper,the overall framework and workflow of the proposed system are described in detail,and the feasibility and validity of the system are verified by experiments.