Statistically Orthogonal Analysis Reaserch On Color Image Biometrics Identification Recognition

Color image face recognition is playing an important role in Biometrics IdentificationTechnology. Compared to the traditional gray image recognition, color image contain moreinformation. However, the correlation between the three principle components of color imageinfluences the effectiveness of face recognition seriously. Therefore, face recognition focus onhow to extract discrimnant features of color image and remove the correlation between the threeprinciple components of color image. This paper proposes three effective color image facerecognition methods.First of all, according to the statistical unrelated correlation discriminant analysis,Statistically Orthogonal Analysis is proposed for color image face recognition. It is the first timethat this theory is applied on feature level, which extract R, G, B three color principlecomponents features and make the three groups features keep statistical uncorrelated at the sametime. Compare to the traditional color image face recognition algorithms, the proposed methodcan extract more effective color image features. Then, Color Image Statistically OrthogonalAnalysis is exptended in kernel space, which can effectively solve inseparable problem in thelow dimensional space, it can also make full use of the information in the high dimension spaceat the same time, thus improve the efficiency of the algorithm. Finally, an Enhanced Color ImageStatistically Orthogonal Analysis method is put forward. By changing the Fisher discriminantioncriteria to the Maximum Scatter Difference Critera (MSDC), it can effectively avoid singularityproblem, thus improve the performance of the algorithm effectively.Expriment results on three color databases: AR color face database, FRGC-v2color facedatabase and PolyU palmprint database demonstrate the efficiency of the three proposedmethods.