Cross-spectral palmprint recognition with low-rank canonical correlation analysis

As an important biometric trait, palmprint has been widely studied in individual identification. With the popularity of palmprint recognition, many palmprint acquisition devices with different spectra have been designed and applied into practical application, so it is difficult to ensure that the spectra of collectors are consistent during training and testing, which leads to the challenge of cross-spectral palmprint images classification. To address this problem, this paper presents a domain adaptive method based on subspace learning, i.e., low-rank canonical correlation analysis (LRCCA). The proposed method seeks to find the common subspace of cross-spectral palmprint images and capture the low-rank structural relationships in data simultaneously. We perform the experiment on the multi-spectral palmprint dataset with 12 cross-spectral palmprint recognition tasks. The experimental results show that our method achieves the promising results in both identification and verification and outperforms the classical transfer learning methods and deep canonical component analysis (DCCA).