Search

Palmprint biometrics has a broad application prospect owing to non-intrusiveness, ease of image acquisition and stable textural pattern. Hand-crafted approaches are vulnerable to non-ideal palmprint images caused by uneven illumination, motion blur and noise contamination. Many researchers have designed excellent texture descriptors or/and advanced image pre-processing algorithms. Nevertheless, they are highly targeted at some specific data, less adaptable to the emerging data. In this paper, a semi-pretrained contactless palmprint recognition deep network is developed to achieve high accuracy and robustness. Semi-CPRN is composed of underlying network structure from ResNet-152 and the proposed assistant layers dedicated to high-performance palmprint recognition. The well-designed assistant layers enhance convolutional neural network to steadily extract the real palmprint features even from the degraded images without being deceived by the degradation factors. Besides, to better carry out the research on palmprint recognition in the open environment, we established a new contactless database HIT-NIST-V1 under natural scene. The comparative experiments on CASIA, IITD, PolyU3D/2D, Tongji and HIT-NIST-V1 illustrate that Semi-CPRN is comparable and superior to previously published state-of-the-art approaches. Simultaneously, CNN-based palmprint biometrics methods show significant robustness to motion blur, Gaussian noise, and salt and pepper noise. [ABSTRACT FROM AUTHOR]

Finger-vein recognition has been widely explored for biometrics security over the past two or three decade. Its size and characteristics against presentation attacks make it suitable for various commercial, governmental and forensics applications. As with other biometric traits, finger-vein recognition approaches mainly depend on well-engineering feature descriptors and recently, more deep learning (DL) models have been designed to achieve high-performance biometrics. However, the existing DL approaches usually employ the neural networks with increasing layers and parameters, which inevitably leads to an increase in memory consumption and algorithm complexity. Besides, all the vein recognition networks use square filters, which are influenced by network designed for object detection and classification. In this paper, two network design issues for finger-vein recognition are explored: shallow network using rectangular filters and lightweight semi-pretrained network. The exhaustive experiments on three benchmark databases HKPU, FV-USM and SDUMLA demonstrate that rectangular filters outperform square filters for DL-based finger-vein recognition, and the lightweight semi-pretrained network also outperforms non-pretrained network. [ABSTRACT FROM AUTHOR]