Logistic regression classifier for palmprint verification

We propose a supervised machine learning approach for automatic palmprint verification. In our approach a pair of palmprint images is represented and characterized using a vector of regional similarity features. Every regional similarity feature is computed using local modified complex wavelet structural similarity indexes (CW-SSIM). The logistic regression classifier verifies whether two palmprints described by the feature vector belong to same person or not. The aim of our classifier is to improve the matching accuracy and robustness of the verification, based on learned knowledge about: 1) the local and global characterization of the errors arising due to inaccurate image registration (translations, rotations, and distortions), and 2) the underlying vector patterns of the two palmprint images. Our experimental results show that the proposed approach achieves high verification accuracy.