Research Of Image Correlation Analysis For Biometric Identification

Biometrics is an emerging technology that utilizes distinct physiological or behavioral characteristics to verify the identity of an individual. Thus, it is a critical problem to protect the security and integrity of the biometric data for ensuring valid biometric identification. Recently, correlation analysis methods, making use of correlation between biometric images and cover images, become popular to protect biometric data.In this paper, first I introduce the research and development of correlation analysis and hiding techniques applied for protecting biometrics information. Then, correlation analysis related mathematic theory and some fundamental knowledge about information hiding are presented briefly. Afterwards, two different correlation analysis algorithms are proposed in this paper. One is regression method (least absolute shrinkage and selection operator, least angle regression and so on) using the biometric images directly instead of the features of the images, and this method yielding lower residual is faster than former methods. The other one is improvements on optimally pruned extreme learning machine (OPELM), which is the first time used for correlation analysis and performs well. However, considering its time consuming, here we put forward two quick OPELMs to enhance it, one is improving the ranking method, and the other one is changing the construction matrix. Moreover, face images can get low residuals using above methods, which cannot be implemented in previous methods, such as genetic algorithm (GA) combined with principle component analysis (PCA). After getting the residuals between the biometric images and cover images, discrete wavelet transform (DWT) combined with human visual system (HVS) model is adopted to hide and extract the residuals to prove the effectiveness of the proposed methods.In this paper, the biometric images are from the ORL Database of Faces, the Yale Face Database, the Extended Yale Face Database B, PolyU Palmprint Database and CASIA Iris Image Database. I do extensive comparisons experiments between the proposed correlation analysis methods and the algorithms presented on the platform of MATLAB. All the results demonstrate that our proposed technologies are more robust, faster and lower residuals than former methods.