Multispectral Palmprint Recognition Based on Multidirectional Transform

Abstract

Multispectral palmprint recognition is one of the most useful biometric techniques due to features obtained from different spectral resolutions/wavelengths. In this paper, we propose a multidirectional transform-based feature encoding plan for reliable and robust representation and matching of multispectral palm images. The method extracts the region of interest (ROI) for palmprint images captured with non-contact sensors. The registered ROI of each band is newly downsampled using DWT. This approach allows us to take more lines into consideration for interpolation. A undecimated dual-tree complex wavelet transform based multidirectional feature encoding plan is then newly applied since it provides better shift invariance and directional selectivity. Finally, a binary code matching strategy with score level fusion is used to compute matching for efficient identification. The experimental results obtained on CASIA and PolyU datasets show that the presented method gives better results in the blurring binary code matching case than state-of-the-art methods and provides comparable performance in the non-blurring binary code matching.