Elsevier

Neurocomputing

Volume 71, Issues 13–15, August 2008, Pages 3032-3036
Neurocomputing

Letters
Palmprint recognition using Gabor feature-based (2D)2PCA

https://doi.org/10.1016/j.neucom.2007.12.030Get rights and content

Abstract

In this paper, we propose a novel approach of Gabor feature-based (2D)2PCA (GB(2D)2PCA) for palmprint recognition. Three main steps are involved in the proposed GB(2D)2PCA: (i) Gabor features of different scales and orientations are extracted by the convolution of Gabor filter bank and the original gray images; (ii) (2D)2PCA is then applied for dimensionality reduction of the feature space in both row and column directions; and (iii) Euclidean distance and the nearest neighbor classifier are finally used for classification. The method is not only robust to illumination and rotation, but also efficient in feature matching. Experimental results demonstrate the effectiveness of our proposed GB(2D)2PCA in both accuracy and speed.

Introduction

Palmprint recognition has attracted wide attention from researchers as a new biometrics authentication technology. The palmprint, the unique inner surface of hand, contains a number of distinctive features such as principal lines, wrinkles, ridges and minutiae. Compared with other technologies, the advantages of palmprint recognition are low resolution, low cost, non-intrusiveness and stable structure features [2].

Studies on palmprint recognition have focused on the feature extraction of line, texture, statistics and multiple representations. Line-based recognition is often seen in earlier works on off-line palmprint images with high resolution (up to 500 dpi) [14]. But for online palmprint images, the line feature extraction is difficult for the low-resolution images (less than 100 dpi). Texture-based features can be obtained by transform and wavelet [8]. 2D Gabor filter is a common texture-based method for palmrprint recognition [13], [5], which provides high-recognition accuracy at the cost of speed and memory. Statistical approaches, such as Principal Component Analysis (PCA), linear discriminative analysis (LDA) and two-dimensional PCA (2DPCA), obtain eigenpalms, fisherpalms and some other features in palmprint recognition [9], [11], [7]. The fusion of different features and the integration of multiple approaches are also used in the late literature [4], [6], which lead to relatively satisfying results.

PCA is a representative statistical method, which can effectively reduce the dimension of image space as a whole, whereas 2DPCA treats the images directly with a better recognition performance and time consumption [12]. However, the main disadvantage of 2DPCA is that too many coefficients are needed for image representation. (2D)2PCA, an improved 2DPCA algorithm, overcomes the problem by projecting the images onto row and column directions simultaneously [15].

Nevertheless, either PCA or its derived methods are sensitive to variations caused by illumination and rotation. But some works [13], [5] showed that Gabor filters can provide robust features against varying brightness and contrast of images. In their method, Gabor features, derived from the convolution of a Gabor filter and palmprint images, were represented and matched by hamming code and hamming distance, respectively. However, the procedure for feature coding and matching by pixels requires too much time and memory. Moreover, to extract more local features from the original images, a series of Gabor filters with various scales and orientations (called Gabor filter bank) are needed in most cases of biometrics. Eventually this will enlarge the feature dimension by time and make feature matching beyond implementation.

Taking together both procedures, we develop a new algorithm for palmprint recognition, Gabor feature-based (2D)2PCA (GB(2D)2PCA). The proposed GB(2D)2PCA consists of three steps: initially, Gabor feature matrix of different scales and orientations are extracted by the convolution of Gabor filter bank and original images. All the Gabor feature matrices of training samples compose Gabor feature space. Subsequently, (2D)2PCA reduce the dimension of Gabor feature space in both row and column directions resulting in fewer coefficients for feature matching. Finally, Euclidean distance and the nearest neighbor classifier are used for feature matching and classification.

The rest of this paper is organized as follows: 2 Gabor filter bank for feature extraction, 3 (2D) present our algorithm in sequence, the former introduces Gabor filter bank for feature extraction, and the latter explains (2D)2PCA for dimensionality reduction of feature space. Experimental results are given in Section 4. Section 5 highlights the conclusions.

Section snippets

Gabor filter bank for feature extraction

2D Gabor has the following general form [1]:G(x,y,θ,u,σ)=12πσ2exp{-x2+y22σ2}exp{2πi(uxcosθ+uysinθ)}where i=-1, u is the frequency of the sinusoidal wave, θ controls the orientation of the function and σ is the standard deviation of the Gaussian envelope. Kong et al. [5] designed 12 Gabor filter combinations of different u, θ and σ. By comparing the receiver operating characteristic (ROC) curves, θ=0, u=0.0916 and σ=5.6179 is the favorite group for 128×128 palmprint images.

Taking into account

(2D)2PCA for dimensionality reduction of Gabor feature space

In 2DPCA, the covariance matrix G can be evaluated by G=1Ni=1N(Xi-X¯)T(Xi-X¯)whereX¯=1NiXi.

Since the size of Xi is 960×32, G has a dimension of 32×32. The orthonormal eigenvectors of G corresponding to the d largest optimal value is proved to be optimal projection matrix [3]Ropt=[r1,,rd].

The value of d can be determined by the ratio of the sum of chosen d largest eigenvalues to all.

Similarly, the optimal projection in the column direction Copt=[c1,,cq] is obtained by the transposed space

Palmprint database

We collected 800 left-hand images of 595×790 pixels of 75 dpi resolution from 80 subjects by a digital scanner in our lab at an interval of 3 month. The central 128×128 pixels of each hand image extracted by the preprocessing method [10] constitute a palmprint database (Fig. 1). Five images of each subject are randomly chosen for training, and the remaining five images are used for testing. So the training set and testing set contains 400 images, respectively. Fig. 2 exemplifies some palmprint

Conclusion

This paper reports a novel GB(2D)2PCA method for palmprint recognition. The novelty of the GB(2D)2PCA comes from implementation of(2D)2PCA on an augmented Gabor feature vector first derived from a Gabor filter bank, not a single Gabor filter, for palmprint images. The GB(2D)2PCA method, which is more robust to variations of illumination and rotation, uses Gabor feature vector of five scales and six orientations as an input of (2D)2PCA instead of raw palmprint images. Meanwhile, GB(2D)2PCA can

Acknowledgments

The authors are grateful to Dr. Lucheng Cao and the anonymous reviewers for their constructive comments and advices. This work is supported partly by the National Natural Science Foundation of China under Grant Nos. 60472033, 60672062, and the National Grand Fundamental Research 973 Program of China under Grant No. 2004CB318005.

Xin Pan received her B.S. and M.S in Xi’an Institute of Technology and Inner Mongolia Agricultural University in 1997 and 2000. She has worked in College of Computer and Information Engineering, Inner Mongolia Agricultural University since then. Now she is pursuing her Ph.D. degree at the Institute of Information Science, Beijing Jiaotong University. Her research interests include image processing, pattern recognition, etc.

References (15)

There are more references available in the full text version of this article.

Cited by (73)

  • Learning 2D Gabor filters by infinite kernel learning regression

    2021, Journal of Computational Mathematics and Data Science
  • Multiple feature fusion for unconstrained palm print authentication

    2018, Computers and Electrical Engineering
    Citation Excerpt :

    This issue increases the memory requirements and takes more time to evaluate feature sets. To overwhelm the local information loss, large coefficient requirement and to get same or increased accuracy level, 2D2PCA has been applied by projecting 2DPCA onto row and column directions simultaneously [26]. However, it is suspected that some of the original extracted features may be noisy or not relevant and the addition of these features may lead to larger computations and deterioration in the classification accuracy.

  • Palmprint recognition with Local Micro-structure Tetra Pattern

    2017, Pattern Recognition
    Citation Excerpt :

    In addition, the well-known 2D matrices-based subspace analysis is also used in palmprint feature representation including 2D PCA, 2D LDA [12], 2D PCA+PCA [13], and 2D locality preserving projecting (2D LPP) [14]. Furthermore, in order to enhance the discriminative capability, the Fourier Transform [15,16], Discrete Cosine Transform (DCT) [17], Wavelet Transform [18], and Gabor Transform [19,20] were performed along with aforementioned statistical techniques on the palmprint images to extract palm features. However, these approaches require well-aligned palmprint images, and are very sensitive to illumination, distortion, translation, and rotation variances.

View all citing articles on Scopus

Xin Pan received her B.S. and M.S in Xi’an Institute of Technology and Inner Mongolia Agricultural University in 1997 and 2000. She has worked in College of Computer and Information Engineering, Inner Mongolia Agricultural University since then. Now she is pursuing her Ph.D. degree at the Institute of Information Science, Beijing Jiaotong University. Her research interests include image processing, pattern recognition, etc.

Qiu-Qi Ruan was born in 1944. He received the B.S. and M.S. degrees from Northern Jiaotong University, China in 1969 and 1981, respectively. From January 1987 to May 1990, he was a visiting scholar in the University of Pittsburgh, and the University of Cincinnati. Subsequently, he has been a visiting professor in USA for several times. He has published 2 books and more than 100 papers, and achieved a national patent. Now he is a professor, doctorate supervisor at the Institute of Information Science, Beijing Jiaotong University. He is a senior member of IEEE. His main research interests include digital signal processing, computer vision, pattern recognition and virtual reality.

View full text