Palm Vein Recognition Algorithm using Multilobe Differential Filters

Abstract

In this article the new algorithm for palm vein recognition using multilobe differential filters is proposed. After palm vein image preprocessing vein structure is detected based on principal curvatures. The image is considered as a surface in a three-dimensional space. Some vein points are selected using the maximum principal curvature values, and the other vein points are found from starting points by moving along the direction of minimum principal curvature. Multilobe differential filters are used to extract feature maps for vein images. These filters are flexible in terms of basic lobe choice and spatial configuration of lobes. The multilobe differential filters used in the article simulate vein branch points, and Gaussian kernel is used as the basic lobe. The normalized root-mean-square error is applied for image matching. Experimental results using CASIA multi-spectral palmprint image database demonstrate the effectiveness of the proposed method. The value of EER=0.01693 is obtained.