Research On Face Recognition And Bi-module Biometric Key Generation Algorithm Using Sub-pattern LBP Methods

Biometric Encryption introduces biometrics into cryptographic architecture to enhance the security of the encryption system. Of all the biometrics, face and fingerprint are much more popular than the others due to their collectability and acceptability. Their theoretical and practical values become the focus of research efforts in this filed.The local binary pattern (LBP) and its extensions, such as Gabor-LBP andε-LBP, are both derived from a general definition of texture in a local neighborhood. Recently, reseachers use LBP histogram as the feature description of face successfully. For fingerprint, we commonly use Gabor-LBP as its feature extracting method due to the compact distance of inter-calss fingerprint samples. In the end, we present the Bi-module Biometric Key Generation Algorithm based on sub-pattern LBP methods which combines two biometric traits-face and fingerprint with corresponding user id.The primary work of this paper can be summarized as follow:1. We first dicuss the sub-pattern methods of LBP and its entensions, then present a new local classifier framework based on voting. Within this framework, we implement a method called Sub-pattern LBP based on Voting (V-LBP). V-LBP votes all the local regions'labels to get the final label of the entire image. Comparison experiments on occlusion existed AR facial database indicate that each of the two sub-pattern methods-V-LBPH and Sub-pattern LBP based on feature concatenating has superiority. Though V-LBP can get more superiority when there exist occlusions in images.2. We present a new facial classifier called Local Ridge Regression based on Gabor-LBP (GLBP-LRR). GLBP-LRR makes up the insufficient feature extraction of LRR, and can make more accurate voting when there are fewer regions. In order to imrove voting accuracy, we add region detection before GLBP-LRR, thus forming GLBP-LRR base on Region Selection Classifier (SGLBP-LRR). Both the two presented classifiers are tested on AR database, and their classification accuracy are all improved relative to their former and other sub-pattern methods, such as Aw-SpPCA and SpCCA.3. On the basis of these studies, we present Bi-module Biometric Key Generation Algorithm which based on sub-pattern methods of LBP and its extensions. The new algorithm contains four modules: feature extraction module, BioHash key generation module, key match module and decision module. The final generated key is used as the input key of AES cryptosystem. Experiments on FVC2000 and FVC2002 fingerprint database and ORL facial database indicate that the presented algorithm develops in accuracy and security relative to single biometric key generation algorithm.