Robust iris recognition using sparse error correction model and discriminative dictionary learning

doi:10.1016/j.neucom.2013.06.051

Neurocomputing

Volume 137, 5 August 2014, Pages 198-204

https://doi.org/10.1016/j.neucom.2013.06.051 Get rights and content

Abstract

Robust iris recognition is a hot research topic in the biometrics community and the sparse representation-based methods are promising to achieve desirable robustness and accuracy. Motivated by the fact that corruptions and occlusions incurred by eyelash occlusions, eyelid overlapping, specular and cast reflection in iris images are spatially localized but large in magnitude, we present a robust iris recognition method based on a sparse error correction model. In the proposed method, all the training images are concatenated as a dictionary and the iris recognition task is cast to an optimization problem to seek a sparse representation of the test sample in terms of the dictionary. And a sparse error correction term is introduced into the objective function of the optimization problem to deal with gross and spatially localized errors. Furthermore, in order to compact the huge dictionary, we introduce a discriminative dictionary learning framework to reduce computational complexity. Experimental results on CASIA Iris Image Database V3.0 show that the proposed methods achieve competitive performance in both recognition accuracy and efficiency.

Introduction

Iris recognition is one of the most reliable and secure methods for personal identification and authentication [1], which has been attracted an increasing attention in the past two decades. Two most well-known and representative iris recognition frameworks were developed by Daugman [2] and Wildes [3]. Daugman [2] applied a 2-D Gabor filter to extract the features from the scale-normalized iris images and quantized them to 256 bytes iris codes. Then, the normalized Hamming distance between the iris codes was employed as the measure for recognition. In contrast, Wildes [3] used convolution with Laplacian of Gaussian filter at multiple scales to produce a template of the iris texture and then computes the normalized correlation as the similarity metric. Motivated by their pioneer works, a lot of iris recognition algorithms have been presented to pursue more excellent performance. The earlier works were summarized in [1] and some recent efforts were reported in [5], [6], [7], [8], [9], [10]. Existing iris recognition algorithms based on feature extraction and matching have achieved very high recognition accuracy for clean iris images [4].

However, the quality of the acquired iris images is often poor in practical scenarios due to various side effects such as motion blur, defocus blur, eyelid overlapping, specular and cast reflection. This leads to heavy performance degradation for most existing methods so far [1], [5]. Improving the robustness of the iris recognition algorithm is still a challenging issue in the iris biometric community. Si et al. [8] focused on the robustness of iris segmentation and developed an eyelash detection algorithm based on directional filters to enhance the iris segmentation, achieving promising results. Li and Ma [9] presented a robust method based on the Random Sample Consensus (RANSAC) to localize the non-circular iris boundaries and an image registration method based on the optical flow algorithm to account for iris image deformation. Instead of the traditional feature extraction and matching framework, Pillai et al. [10] proposed a novel framework based on random projection and sparse representation-based classification (SRC) [11]. This method achieves desirable robustness to blur, segmentation error and small level occlusions. However, when the level of occlusions is relatively high, the recognition rate of the algorithm decreases remarkably. Moreover, the SRC algorithm requires a sufficiently large number of training images from each class to span subspace of each class and stacks all training images to form a dictionary directly, which leads to a large size of dictionary and intensive computation complexity.

In this paper, we firstly present an improved SRC iris recognition method based on sparse error correction (SEC) mode [11] to improve the robustness of the recognition algorithm to the high level occlusions and corruptions. The extra robustness derived from the SEC mode comes from the presence of a sparse error term which corresponds to the occlusions and corruptions of the iris image. Since the errors incurred by occlusions, secularities and cast shadows are large in magnitude but sparse in space, it is more reasonable that a test sample is better represented by a combination of the training samples with additive sparse and large-magnitude errors than by the standard SRC model used in [10]. Furthermore, a discriminative dictionary learning method for sparse error correction model (SEC-DKSVD) derived from discriminative K-SVD (D-KSVD) [12] is introduced to compact dictionary to improve both the efficiency and accuracy of the iris recognition algorithm. Finally, in order to detect and reject those invalid test images, we propose a validation scheme based on the cumulative sparsity concentration index (CSCI) [10]. Differing from [11], [10] in which SCI of the recovered sparse representation coefficient vector was applied as the confidence of the recognition result, the suggested CSCI-based strategy employs SCI of the label vector obtained by the learned classifier. Besides the ability of detecting and rejecting the invalid samples, this scheme improves the robustness of iris recognition as well.

The rest of the paper is organized as follows. In Section 2, the adaptability of the SR-based method for iris recognition task is discussed, and the proposed sparse error correction model for iris recognition is presented. In Section 3, the improved discriminative dictionary learning method for error correction model is introduced. In Section 4, the validation scheme for recognition results is suggested. Experimental results are reported in Section 5, and we conclude the paper in Section 6.

Section snippets

Sparse error correction model for iris recognition

Recently, Wright et al. [11] reported a powerful tool, the SRC algorithm, for face recognition. SRC depends on low-dimensional linear models for illumination variation in recognition tasks. The SRC algorithm relies on low-dimensional linear models for illumination variation in recognition tasks. More specifically, given n training samples $d_{i, 1}, d_{i, 2}, \dots, d_{i, n} \in R^{m}$ of the ith class object, the new test sample $y_{i} \in R^{m}$ of this object will lie near the linear span of the training samples $y_{i} = d_{i, 1} x_{i, 1} + d_{i, 2} x_{i, 2}$

D-KSVD for sparse error correction model

The sparse error correction model-based iris recognition algorithm uses the entire iris images training set as a dictionary to classify a new test sample, which may result in a huge size dictionary as the basic SRC method and expensive computation. Therefore, shrinking the size of the dictionary becomes a practical issue. The K-SVD algorithm proposed in [21] addressed this issue by learning a compact dictionary from the training set with N items ( $N < n$ ) for each class by solving the following

Validation scheme based on CSCI

In a practical recognition system, a new test sample could be an iris image of a subject that is not authorized or an image that is not an iris at all. Thus, the ability to detect and reject suspect recognition results and invalid test samples is crucial for an iris recognition system. The validity of the test sample is implicated in the distribution of the coefficient vector and the label vector given by (12). The notion of sparsity concentration index (SCI) was introduced in [11] to validate

Experimental results

In this section, we evaluate the proposed recognition methods on CASIA-IrisV3 [14]. There are three subsets in CASIA-IrisV3, CASIA-Iris-Interval, CASIA-Iris-Lamp and CASIA-Iris-Twins, containing a total of 22,034 iris images from more than 700 subjects and 1500 classes (eyes) acquired by CASIA close-up iris cameras or OKI IRISPASS-h hand-held iris sensors. To evaluate robustness of recognition algorithms to elastic deformation of iris texture, CASIA-Iris-Lamp subset collected under different

Conclusion

In this paper, we introduced a SEC mode-based method and a SEC-DKSVD-based method for the iris recognition. Furthermore, a CSCI-based results validation strategy employing the sparsity of the label vector obtained by the classifier learned by SEC-DKSVD was proposed. Our experiments on CASIA-Iris-Lamp indicate that the proposed methods show competitive performance when a sufficiently large number of training samples are available. On one hand, under the same conditions, the proposed SEC-based

Acknowledgment

This task is supported by the Nation Natural Science Foundation of China (Grant nos. 61001123, 60973113); Hunan Province Science and Technology Planning Project (Nos. 2013FJ4033, 2009FJ3005, 2012FJ4129, 2012GK3056); the Science Research Key Project of the Education Department of Hunan Province (No. 13A107); and Changsha Science and Technology Planning Key Project (Nos. K1104022-11, K1207027-11).

Yun Song received his M.S. in computer application from Changsha University of Science & Technology, China in 2008. He is currently is an associate professor in the School of Computer and Communication Engineering of Changsha University of Science & Technology, China. And he is pursuing his PhD from Hunan University, China. His major research interests include video coding, multimedia communication and security, compressive video sensing, video analysis, and image processing.

References (21)

K.W. Bowyer et al.
Image understanding for iris biometricsa survey
Comput. Vis. Image Underst.
(2008)
P. Li et al.
Iris recognition in non-ideal imaging conditions
Pattern Recognit. Lett.
(2012)
J. Daugman
High confidence visual recognition of persons by a test of statistical independence
IEEE Trans. Pattern Anal. Mach. Intell.
(1993)
R.P. Wildes
Iris recognitionan emerging biometric technology
Proc. IEEE
(1997)
J.G. Daugman
Probing the uniqueness and randomness of iris codesresults from 200 billion iris pair comparisons
Proc. IEEE
(2006)
E.M. Newton et al.
Meta-analysis of third-party evaluations of iris recognition
IEEE Trans. Syst. Man Cybern.
(2009)
K.P. Hollingsworth et al.
Improved Iris recognition through fusion of Hamming distance and fragile bit distance
IEEE Trans. Pattern Anal. Mach. Intell.
(2011)
V.D. Sunita
Robust iris recognition based on statistical properties of Walsh Hadamard transform domain
Int. J. Comput. Sci. Issues
(2012)
Y. Si et al.
Novel approaches to improve robustness, accuracy and rapidity of iris recognition systems
IEEE Trans. Ind. Inf.
(2012)
J.K. Pillai et al.
Secure and robust iris recognition using random projections and sparse representations
IEEE Trans. Pattern Anal. Mach. Intell.
(2011)

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

Cited by (21)

Collaborative representation based face classification exploiting block weighted LBP and analysis dictionary learning
2019, Pattern Recognition
Citation Excerpt :
Zhang et al. [19] proposed a discriminative K-SVD method for face recognition. In contrast to conventional sparse representation based classification methods in which a dictionary is a set of pre-defined samples, the learning of a dictionary is able to adapt its contents to a given set of signal examples can provide better performance for image classification [19–21]. In general, dictionary learning approaches can be categorized into two types, synthesis and analysis dictionary learning, according to the way of coding an input signal during the learning stage [3,22,23].
Traditional collaborative representation based classification (CRC) method usually faces the challenge of data uncertainty hence results in poor performance, especially in the presence of appearance variations in pose, expression and illumination. To overcome this issue, this paper presents a CRC-based face classification method by jointly using block weighted LBP and analysis dictionary learning. To this end, we first design a block weighted LBP histogram algorithm to form a set of local histogram-based feature vectors instead of using raw images. By this means we are able to effectively decrease data redundancy and uncertainty derived from image noises and appearance variations. Second, we adopt an analysis dictionary learning model as the projection transform to construct an analysis subspace, in which a new sample is characterized with the improved sparsity of its reconstruction coefficient vector. The crucial role of the analysis dictionary learning method in CRC is revealed by its capacity of the collaborative representation in an analytic coefficient space. Extensive experimental results conducted on a set of well-known face databases demonstrate the merits of the proposed method.
Fisher discrimination dictionary pair learning for image classification
2017, Neurocomputing
Citation Excerpt :
When the learned dictionary is universal to all classes, discrimination of universality representation can be explored by using group sparsity [30] on coding coefficients. Based on [6], Song et al. [8] adopted sparse error correction (SEC) model to minimize the energy of the errors and proposed a discriminative dictionary learning model for iris recognition robust to corruption and occlusion. By introducing Fisher criterion, Zheng and Tao [9] proposed Fisher discriminative KSVD (FD-KSVD) to learn an over complete discriminative dictionary and an optimal linear classifier for image classification; Xu et al. [34] proposed Fisher discriminant unsupervised dictionary learning for improving the clustering performance of current dictionary learning approaches in unsupervised scenarios; and Dong et al. [11] proposed a supervised dictionary learning algorithm based on local Fisher discrimination for action recognition in still images.
Dictionary learning has played an important role in the success of sparse representation. Although several dictionary learning approaches have been developed for image classification, discriminative dictionary pair learning, i.e., jointly learning a synthesis dictionary and an analysis dictionary, is still in its infant stage. In this paper, we proposed a novel model of Fisher discrimination dictionary pair learning (FDDPL), in which Fisher discrimination information is embedded into analysis representation, analysis dictionary, and synthesis dictionary representation. With the proposed Fisher-like discrimination term, discrimination of both synthesis dictionary representation and analysis dictionary representation is introduced into the dictionary pair learning model. An iterative algorithm to efficiently solve the proposed FDDPL and a FDDPL based classifier are also presented in this paper. The experiments on face recognition, scene categorization, gender classification, and action recognition clearly show the advantage of the proposed FDDPL.
Synthesis linear classifier based analysis dictionary learning for pattern classification
2017, Neurocomputing
Citation Excerpt :
SR represents a sample by the linear combination of several dictionary’s atoms chosen out of the whole dataset (e.g., the classic sparse representation based classification [1]) or other predefined bases [9,10], so the dictionary is crucial in the sample reconstruction process. Learning an optimal dictionary from data instead of using a set of predefined bases has achieved a big success in various practical tasks, such as face recognition [11], image classification [12], image and speech denoising [13,14], and others [15]. According to the ways of encoding input data, dictionary learning (DL) approaches [16] can be mainly classified as synthesis dictionary learning (SDL) and analysis dictionary learning (ADL).
Dictionary learning approaches have been widely applied to solve pattern classification problems and have achieved promising performance. However, most of works aim to learn a discriminative synthesis dictionary and sparse coding coefficients for classification. Until recent years, analysis dictionary learning began to attract interest from researchers. In this paper, we present a novel discriminative analysis dictionary learning frame, named Synthesis Linear Classifier based Analysis Dictionary Learning (SLC-ADL). Firstly, we incorporate a synthesis-linear-classifier-based error term into the basic analysis dictionary learning model, whose classification performance is obviously improved by making full use of the label information. Then, we develop an alternating iterative algorithm to solve the new model and obtain closed-form solutions leading to pretty competitive running efficiency. What is more, we design three classification schemes by fully exploiting the synthesis linear classifier. Finally, extensive comparison experiments on scene categorization, object classification, action recognition and face recognition clearly verify the classification performance of the proposed algorithm.
Discriminative analysis-synthesis dictionary learning for image classification
2017, Neurocomputing
Citation Excerpt :
With the theoretical development of synthesis and analysis dictionary learning, dictionary learning based sparse representation has been widely applied to image denoising [1], face recognition [7,8], image classification [9,10] and detection [5]. For example, Song et al. [22] adopted sparse error correction (SEC) model to minimize the energy of the errors and proposed SEC-DKSVD for iris recognition. By introducing Fisher criterion, Zheng et al. [23] proposed Fisher discriminative KSVD (FD-KSVD) to learn an over complete discriminative dictionary and an optimal linear classifier for image classification and Dong et al. [24] proposed a supervised dictionary learning method for action recognition in still images.
Dictionary learning has played an important role in the success of sparse representation. Although discriminative synthesis dictionary learning for sparse representation with a high-computational-complexity l₀ or l₁ norm constraint has been well studied for image classification, jointly and discriminatively learning an analysis dictionary and a synthesis dictionary is still in its infant stage. As a dual of synthesis dictionary, the recently developed analysis dictionary can provide a complementary view of data representation, which can have a much lower time complexity than sparse synthesis representation. Although several class-specific analysis-synthesis dictionary, which may have a big correlation between different classes' dictionaries, have been developed, how to learn a more compact and discriminative universal analysis-synthesis dictionary is still open. In this paper, to provide a more complete view of discriminative data representation, we propose a novel model of discriminative analysis-synthesis dictionary learning (DASDL), in which a linear classifier based on the coding coefficient is jointly learned with the dictionary pair, thus the performance of the classifier and the representational power of the dictionary pair being considered at the same time by the same optimization procedure. The size of the learned dictionaries can be very small since the analysis-synthesis dictionary is shared by all class data. An iterative algorithm to efficiently solve the proposed DASDL is presented in this paper. The experiments on face recognition, gender classification, action recognition and image classification clearly show the superiority of the proposed DASDL.
Learning a structure adaptive dictionary for sparse representation based classification
2016, Neurocomputing
Citation Excerpt :
In the first category, a dictionary shared by all classes and a classifier trained on the representation coefficients computed on the dictionary are learnt jointly [5,6,10–13]. Based on [6], Song et al. [14] adopted sparse error correction (SEC) model to minimize the energy of the errors and proposed SEC-DKSVD for iris recognition. By introducing Fisher criterion, Zheng and Tao [15] proposed Fisher discriminative KSVD (FD-KSVD) to learn an over complete discriminative dictionary and an optimal linear classifier for image classification and Dong et al. [16] proposed a supervised dictionary learning algorithm for action recognition in still images.
Dictionary learning (DL), playing a key role in the success of sparse representation, has led to state-of-the-art results in image classification tasks. Among the existing supervised dictionary learning methods, the label of each dictionary atom is predefined and fixed, i.e., each dictionary atom is either associated to all classes or assigned to a single class. In this paper, we propose a structure adaptive dictionary learning (SADL) method to learn the relationship between dictionary atoms and classes, which is indicated by a binary association matrix and jointly optimized with the dictionary. The binary association matrix can not only represent class-specific dictionary atoms, but also hyper-class dictionary atoms shared by multiple classes. Furthermore, discrimination is explored by introducing Fisher criterion on coding coefficient and reducing between-class dictionary coherence. The extensive experimental results have shown that the proposed SADL can achieve better performance than previous supervised dictionary learning methods on various classification databases.
Robust discriminative non-negative matrix factorization
2016, Neurocomputing
Traditional non-negative matrix factorization (NMF) is an unsupervised method that represents non-negative data by a part-based dictionary and non-negative codes. Recently, the unsupervised NMF has been extended to discriminative ones for classification problems. However, these discriminative methods may become inefficient when outliers are presented in the data, e.g. mislabeled samples, because the outliers usually deviate from the normal samples in one class and would perturb the discriminative dictionary. In this paper, we propose a novel method, called robust discriminative non-negative matrix factorization (RDNMF), to reduce the effect of outliers and improve the discriminative strength. The RDNMF learns a non-negative dictionary for each class, and each dictionary contains two parts: a discriminative part and an outlier part. The discriminative parts are obtained by minimizing the cosine similarity between classes. The codes on the outlier part are required to be sparse so that most outliers can be modeled by this part, without large influence over the discriminative part.The final dictionary is obtained by concatenating the discriminative parts of all classes, and the non-negative codes for each sample, as well as test sample, are obtained by coding with this dictionary. Experimental comparisons with existing dictionary learning methods on MNIST, PIE, Yale B and ORL demonstrate the effectiveness and robustness of our approach.

View all citing articles on Scopus

Wei Cao received his M.S. in computer application from National University of Defense Technology in 1989, China. He is currently an assistant associate professor in the School of Computer and Communication Engineering of Changsha University of Science & Technology, China. His major research interests include video processing technology, multimedia technology, signal processing, pattern recognition, and image processing.

Zunliang He was born in 1988. He is pursuing his M.S. degree at Computer and Communication Engineering Institute, Changsha University of Science & Technology, Changsha, PR China. His research interests include video analysis and image processing.

View full text

Robust iris recognition using sparse error correction model and discriminative dictionary learning

Abstract

Introduction

Section snippets

Sparse error correction model for iris recognition

D-KSVD for sparse error correction model

Validation scheme based on CSCI

Experimental results

Conclusion

Acknowledgment

Comput. Vis. Image Underst.

Pattern Recognit. Lett.

High confidence visual recognition of persons by a test of statistical independence

IEEE Trans. Pattern Anal. Mach. Intell.

Iris recognitionan emerging biometric technology

Proc. IEEE

Probing the uniqueness and randomness of iris codesresults from 200 billion iris pair comparisons

Proc. IEEE

Meta-analysis of third-party evaluations of iris recognition

IEEE Trans. Syst. Man Cybern.

Improved Iris recognition through fusion of Hamming distance and fragile bit distance

IEEE Trans. Pattern Anal. Mach. Intell.

Robust iris recognition based on statistical properties of Walsh Hadamard transform domain

Int. J. Comput. Sci. Issues

Novel approaches to improve robustness, accuracy and rapidity of iris recognition systems

IEEE Trans. Ind. Inf.

Secure and robust iris recognition using random projections and sparse representations

IEEE Trans. Pattern Anal. Mach. Intell.