Channel Attention-Based Adaptive Near Infrared Spectral Deep Residual Network for Palmprint Image Recognition

The palmprint images contains a multitude of unique characteristics for human identity recognition, making it the most potential and influential research content in biometric projects and some palmprint recognition algorithms have been exploited in recent years. Most of the existing methods are mainly aimed at some specific scenarios such as high-resolution palmprint image capture device. However, these methods are not universal in realistic scenarios and these algorithms cannot meet the constantly changing requirement of diversity applications. So as to fully address these issues, this paper presents a concrete framework for Low-resolution palmprint image recognition named Improved Residual Channel Attention Network(IRCANet).Experiments are carried out on datasets Hong Kong PolyU, our method achieves rank-1 accuracy of 99.63% in 15dpi.