The Research Of Face Recognition Based On The Phase Spectrum Of Fractional Fourier Transform

As one of the hottest research directions in the information technology and humancomputer interaction,face recognition technology has been fully studied and applied to many fields in real society.Compared with other biometric features,face recognition technology is more succinct and efficient in personal authentication,has the significant value and wide application foreground.A large number of recognition algorithms and implementation methods have been produced in recent decades,and some of them has been applied with high recognition precision in the international authoritative face recognition database.However,face recognition technology still has many problems such as lack of illumination,small sample size,low resolution of image acquisition equipment and so on.The application is limited in finance payment system or information privacy protection.Fractional Fourier Transform(FRFT)is a recognition algorithm which can map a gray face image to Fourier domain with high efficiency feature extraction.Compared to traditional theories,FRFT is more universal and flexible.In this paper,a new algorithm based on FRFT with high recognition rate and robustness for face recognition is presented.The algorithm was testified in theory and experiments.First,the current difficulties of multi-sample and single-sample face recognition are analyzed.The algorithm based on fractional Fourier transform(FRFT)to extract facial features is proposed.The basic definition,main properties and numerical calculation of FRFT are introduced.The application of two-dimensional form and face recognition is studied emphatically.The correlation properties of face samples in fractional Fourier domain are analyzed in detail.The fractional Fourier domain phase part is conducive to feature extraction and recognition both in signal analysis and image processing.Then,a novel face recognition method based on multi-order FRFT phase band weighted fusion is proposed in the human face global feature extraction.The FRFT phase spectrum is obtained by using the fractional Fourier transform,which eliminates the redundant information in the sample and extraction the phase feature.By optimizing the information fusion method,the FRFT phase spectrum is divided and merged without image-to-vector conversion and the original sample discriminant feature.The two-dimensional discriminant criterion is proposed by selecting the order parameter using the greedy algorithm,and used to select the order to balance the different order characteristics,and make full use of the different order Fourier domain smoothing information and detailed information.It can effectively reduce the external factors,especially the impact of illumination on the face image,so that the efficiency of algorithm recognition and robustness are enhanced.Finally,a single sample face recognition algorithm based on FRFT phase spectrum multi-manifold feature matching based on only one sample storage or registration in the system is proposed.Characterization of human faces using FRFT phase spectrum can effectively find the essential characteristics of the sample without age influence to improve the robustness of the algorithm to age factors.The algorithm fully uses the geometric information between the local features of the sample and the local phase spectrum,and focuses on the grasp of the details of the image.And the problem that the single training sample cannot estimate the intraday divergence matrix by the traditional algorithm problem can be solved.The localized linear embedding(LLE)algorithm is used to identify the reconfigurable spacing between manifolds.In summary,this paper based on fractional Fourier transform has been studied and improved and proposed a new algorithm in order to solve the existing multi-sample and single-sample face recognition technology in the shortcomings.By comparing the simulation experiments in different face database,herein proposed method has achieved good results,and has the significant value and wide application foreground.