Image correlation analysis for biometric identification
Yan Qing Hu, Jin Zhang, Wenfeng Hang · 2011
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. This paper proposes different correlation analysis algorithms. Optimally pruned extreme learning machine (OP-ELM) is the first time used for correlation analysis. However, considering OP-ELM consuming too much time, here we put forward two methods, named quick OPELM, short of qOPELM, and Original Image Correlation Analysis (OICA). The ranking method is improved in the first method, and based on it we change the Hessian matrix to get OICA. Moreover, face images can get low residuals using above methods, which cannot be implemented in previous method, genetic algorithm (GA) combined with principle component analysis (PCA). After we get the residuals between the biometric images and cover images, discrete wavelet transform (DWT) method combined with human visual system (HVS) model is adopted to hide and extract the residuals. All the results demonstrate that our proposed technologies are faster and lower residuals than former methods.