Multi-modal Face Recognition Based On Digital Fringe Projection

Face recognition is one of the most important means for biometricauthentication. The research of face recognition has been gaining more andmore attention due to its wide range of applications in security surveillance,industry automation and entertainment, etc. After years of development,experts have achieved a lot of important research results. However, due to thevulnerability to texture, pose and illumination variation, the recognition ratevaries a lot. All of these challenges will still be hot topics for the researchers.The proposed method is a combination of2D and3D approach which canbest leverage the strength of both methods. A high-speed digital fringeprojection system projects3phase-shifted sinusoidal fringe patterns onto thesubjects face and the camera captures the distorted fringe patterns from anoff-set angle. The face fringe images are analyzed by the phase shiftingmethod and the Fourier transform method to obtain a spectral representationof3D face. One pose estimator based on SVM is adopted before the facerecognizers to deal with the pose varying challenges in practical use. At last,the spectrum image is combined with the gray-scale image for further evaluation by a linear weighted scoring system.Simulation and experimental results tested on large3-D database FRGC2.0demonstrated that the proposed method has low computational cost andstorage expenses while achieving satisfactory recognition rate under differentposes and illuminations.