Research On Face Blacklist Technology With Scarce Training Data

Face recognition has become an important means of identity authentication because of its unique biological characteristics and the convenience of collection methods.However,the actual samples collected in real life often can not meet the training requirements of neural network in deep learning.Therefore,we focus on the technical issues of face blacklist for solving small samples.This paper first reviews and summarizes the relevant theory of face recognition technology and points out some challenges and difficulties faced by the traditional neural network technology in the face of small sample size.Based on this,this dissertation launches the research on face blacklist technology,The dissertation mainly makes use of the theory of image matching network and transfer learning to solve the problem of face blacklist,which is of great significance both in theory and application.The main research work and innovation of the thesis are as follows.Firstly,this dissertation mainly studies the face recognition technology and related theories in depth learning.By summarizing and comparing the theory of feature extraction such as SIFT,HOG,and so on.And the most recent depth learning techniques,this paper proposes a face feature extraction framework based on the deep learning image matching network model.Then,a face recognition model of image matching network based on migration learning is proposed.Through the image pair matching method,small samples of a few hundred can be converted into tens of thousands of image pairs,so as to increase the number of training samples in the network.At the same time,the introduction of migration learning thinking and theory,the model trained on the source dataset is migrated to the target dataset that we need,which can achieve better results on the target dataset with less training set.Finally,an adaptive face recognition system framework based on incremental SVM is proposed and designed aiming at the problem that the traditional face recognition framework can not effectively update the model and that the amount of data required is too large.The model firstly introduces the incremental SVM as the classifier after feature extraction to reduce the amount of data required for the training model.At the same time,the self-training is continuously self-renewed by the incremental model training method.The model can make use of the data collected by the system for updating and perfection,and continuously optimize itself during use.By designing the adaptive face recognition system framework,the disadvantages of the traditional face recognition system can be avoided,so that the system can be trained incrementally to learn new features,avoid using a large amount of data sets,and at the same time,can make the effect of the system Increase the use of time getting better.