Research On Faceauthentication And Encryption Based On Unconstrained Face Image

Face based authentication has achieved very good performance in theconstrained imaging condition. However, in practical application, the face imagesusually are captured with variation of illumination, pose, expression and occlusion.The authentication performance will decreases a lot for such images. In this paper,we propose the face authentication scheme with illumination and pose variation.Furthermore, We propose a biometric template protection scheme based onhomomorphic encryption, which can ensure the security and the performance ofidentity authentication simultaneously.The detail is as follows:Firstly, we propose a scheme to combining Normalized Structural Descriptor(NSD) with Single Scale Retinex (SSR-NSD) for illumination invariant facerecognition. The SSR algorithm will cause shadows under large illumination variation.and the performance degradation. Facing on this problem, Therefore We proposeNormalized Structural Descriptor(NSD) to deal with images from Single ScaleRetinex (SSR). The experimental results on Yale Face Database Band Extended YaleFace Database B show that SSR-NSD is robust to illumination variation. It could getthe good authentication performance.Secondly, we propose a novel face authentication approach based on affine scaleinvariant feature transform and structural similarity. By analyzing the robustness ofASIFT descriptors to pose variation and occlusion, we find that the ASIFT descriptorsare robust to face images with in-plane rotation or off-plane rotation of25degree.And they are robust to expression variation and occlusion. Therefore, we propose thescheme to combining ASIFT and SSIM. The experimental results on FERET,CMU-PIE and AR face databases show that our approach has good performance forface images with expression variation, constrained pose variation or occlusion usingonly one frontal face image for enrollment.Thirdly, We propose an encryption scheme based on homomorphic encryptionand chaotic scrambling, which can ensure the security and the performance of identityauthentication simultaneously.Finally, based on the VC6.0and SQL2008, we develope the face authenticationsystem.