Fast palmprint identification with multiple templates per subject

Palmprint identification system commonly stores multiple templates for each subject to improve the identification accuracy. The system then recognizes a query palmprint image by searching for its nearest neighbor from all of the templates. When applied on moderate or large scale identification system, it is often necessary to speed up this process. In this paper, to speed up the identification process, we propose to utilize the intrinsic characteristics of the templates of each subject to build a tree, and then perform fast nearest neighbor searching with assistance of the tree structure. Furthermore, we propose a novel method to generate the ‘virtual’ template from all the real templates of each subject. The tree constructed by the virtual template and the real templates can further speed up the identification process. Two representative coding-based methods, competitive code and ordinal code, are adopted to demonstrate the effectiveness of our proposed strategies. Using the Hong Kong PolyU palmprint database (version 2) and a large scale palmprint database, our experimental results show that the proposed method searches for nearest neighbors faster than brute force searching, and the speedup becomes larger when there are more templates per subject in the database. Results also show that our method is very promising for embedded system based moderate scale and PC based large scale identification systems.

► In this paper, we present two strategies to facilitate fast and accurate palmprint identification when multiple templates per subject are available in the database.
► Compared with brute force searching, the speedup of strategy 1 is over 1.5 and the identification accuracy is over 99.95% on HongKong PolyU palmprint database.
► We present a novel method to construct a ‘virtual’ template using all the templates of one subject. With the help of the virtual template, the speedup of strategy 2 is over 1.8 on HongKong PolyU palmprint database. On a large scale database, the speedup of strategy 2 is over 2.2 and the identification accuracy is 99.95%.
► Experimental results show that our method is very promising for embedded system based moderate scale (around 1000 registered subjects) and PC based large scale (more than 10,000 registered subjects) identification systems.