Introduction

This is the website for our paper "3D Palmprint Identification Using Block-wise Features and Collaborative Representation", accepted by IEEE Transactions on Pattern Analysis and Machine Intelligence.

During the past decade, great efforts have been devoted in exploiting the palmprint as a biometric identifier. However, most of the previous works have focused on 2D palmprints. Only quite recently, 3D palmprints have begun to draw attention of researchers. Compared with its 2D counterpart, 3D palmprint has several unique merits, including the robustness to illumination variations, the insensitivity to palm contamination, and a better anti-forgery capability. However, most of the existing 3D palmprint matching methods are designed for one-to-one verification and they are not quite efficient to cope with the one-to-many identification case. In this paper, we aim to fill this gap by proposing a novel efficient 3D palmprint identification scheme. We propose to use a collaborative representation (CR) based framework with l1-norm or l2-norm regularizations for 3D palmprint identification. The effects of different regularization terms have been thoroughly evaluated in experiments. To use the CR-based classification framework, one key issue is how to extract feature vectors from raw range images. To this end, we propose a block-wise feature extraction scheme. We at first divide a 3D palmprint ROI (region of interest) into uniform blocks and extract a histogram of surface types from each block; histograms from all blocks are then concatenated to form a feature vector. Such feature vectors are highly discriminative and are robust to mere misalignment between samples. Extensive experiments conducted on the PolyU benchmark dataset demonstrate that the proposed CR-based framework with an l2-norm regularization term can achieve much better recognition accuracy than the other state-of-the-art methods. More importantly, its computational complexity is extremely low, making it quite suitable for the large-scale identification application.

Source Code

CR_L2 based 3D palmprint classification method can be downloaded here, CR_L2.rar.

CR_L1_Homotopy based 3D palmprint classification method can be downloaded here, CR_L1_Homotopy.rar.

CR_L1_DALM based 3D palmprint classification method can be downloaded here, CR_L1_DALM.rar.

HK PolyU 3D palmprint dataset can be downloaded locally here. Note that the original PolyU 3D palmprint database contains both 2D and 3D versions. We only used the 3D ones.

Evaluation Results

In experiments, we used the PolyU 3D palmprint database. This database contains 8000 samples collected from 400 different palms, belonging to 200 volunteers. Among the volunteers, 136 were male and the other 64 were female. 20 samples from each of these palms were collected in two separated sessions, where 10 samples were captured in each session, respectively. The average time interval between the two sessions was one month.
In the following experiments, we took samples collected at the first session as the gallery set and samples collected at the second session as the probe set. Under such an experimental setting, it is easy to know that for the gallery set, there are 400 classes and for each class there are 10 samples. We use the recognition rate as the performance measure. In addition, the running speed of each competing method will also be evaluated. Experiments were performed on a standard HP Z620 workstation with a 3.2GHZ Intel Xeon E5-1650 CPU and an 8G RAM. The software platform was Matlab R2013b.
In our proposed CR-based 3D palmprint identification framework, the regularization term could be the l₁-norm sparsity term or the l₂-norm term. If the l₂-norm term is used, we refer this method as CR_L2. If the l₁-norm sparsity term is used, we tried different methods to solve the l₁-minimization problem, including Homotopy, FISTA, l1_ls, SpaRSA, DALM. And accordingly, we refer these methods as CR_L1_Homotopy, CR_L1_FISTA, CR_L1_l1_ls, CR_L1_SpaRSA, and CR_L1_DALM, respectively. In order to demonstrate the superiority of the proposed CR-based 3D palmprint identification scheme, the proposed method will be evaluated and compared with the other state-of-the-art methods in this field. They include the MCI (mean curvature image)-based method [1], the GCI (Gaussian curvature image)-based method [1], the ST (surface types)-based method [1], and local correlation (LC)-based method [2]. Given a test sample, the time cost for one identification operation includes the time consumed by the feature extraction and the time consumed by matching the test feature with the gallery feature set. The results are summarized in the following Table.

Reference                

[1] D. Zhang, G. Lu, W. Li, L. Zhang, and N. Luo, “Palmprint recognition using 3-D information,“ IEEE Trans. Systems, Man and Cybernetics, Part C, vol. 39, no. 5, pp. 505–519, Sep. 2009.

[2] D. Zhang, V. Kanhangad, N. Luo, and A. Kumar, “Robust palmprint verification using 2D and 3D features,” Pattern Recognition, vol. 43, no. 1, pp. 358-368, Jan. 2010.

Created on: Jun. 28, 2014

Last update: Jul. 07, 2015