Efficient palmprint biometric identification systems using deep learning and feature selection methods

Abstract Over the past two decades, several studies have paid great attention to biometric palmprint recognition. Recently, most methods in literature adopted deep learning due their high recognition accuracy and capability adapt with different acquisition images. However, high-dimensional data a large number of uncorrelated redundant features remain challenge computational complexity issues. Feature selection is process selecting subset relevant features, which aims decrease dimensionality, reduce running time, improve accuracy. In this paper, we propose efficient unimodal multimodal systems based on feature selection. Our approach called simplified PalmNet–Gabor concentrates improvement PalmNet for fast multispectral contactless Therefore, used Log-Gabor filters preprocessing increase contrast features. Then, reduced using dimensionality reduction procedures. For system, fused modalities at matching score level system performance. The proposed method effectively improves reduces as well time. We validated four public databases, CASIA PolyU, Tongji PolyU 2D/3D. Experiments show that our achieves rate while substantially lower