With the rapid development of related technologies in the field of modern biology and the advancement of information technology,the technology of identity verification has slowly shifted to the recognition based on biometrics.Among them,human face,as the most important research direction in the field of biometric identification,has received more and more attention.Different from traditional identification methods,today's identification technology relies mainly on computers and high-tech means to achieve identity determination,such as DNA,through the unique characteristics of the individual to be identified.The human face is an external essential attribute of the human body,and its advantages are self-evident.However,in practical applications,the patterns that are displayed when the face changes in illumination,expression changes,position changes,and occlusion are completely different.This requires a lot of training photos to eliminate these effects.However,in experiments and applications,it is often impossible to obtain a large number of training samples for training to extract features for classification and recognition.This is because the face verification system has very limited storage space and cannot accommodate a large number of training samples.On the other hand,some systems are unable to acquire multiple face sample photos for the same object for training in a short time.For example,in the public security identity verification system,each person in the face sample library has only one photo.Limited training samples do not fully cover all the information needed for face recognition,which is a small sample training problem.Existing algorithms are difficult to improve the accuracy of recognition in small sample training,and in some cases even fail.This paper has made useful exploration and research on the above issues.First-ly,it studies and analyzes some existing classical recognition algorithms in the field of face recognition,studies its algorithm principle and analyzes its mathematical deriva-tion.Research a widely used classification recognition algorithm,sparse classification(SRC)and its derived algorithms.Aiming at the problem that the linear discriminant analysis(FLDA)can not be used in the single-sample case,a new solution based on the original training sample image processing to generate the mirror virtual face is pro-posed,and the SRC classification algorithm is combined to realize the final recognition.Simulation experiments show that the proposed algorithm can not only solve the prob-lem that linear discriminant analysis FLDA can not be used in single-sample case,but also achieve high recognition rate.At the same time,in order to solve the small sample training problem,combined with the advantages of virtual samples.Based on the idea of pixel mapping,two effective algorithms for generating virtual training samples by pixel processing are proposed,and the mathematical calculation process and algorithm prin-ciple of virtual sample generation are analyzed.Combined with sparse classification algorithm,the final recognition and classification are realized.The simulation results show that the proposed algorithm can effectively eliminate the influence of illumination changes when constructing virtual samples,and achieve efficient recognition rate. |