Fusion of Iris and Retina Using Rank-Level Fusion Approach

Transcription

1 Fusion of and Using Rank-Level Fusion Approach A. Kavitha Research Scholar PSGR Krishnammal College for Women Bharathiar University Coimbatore Tamilnadu India N. Radha Assistant Professor GR Govindarajulu School of Computer Technology PSGR Krishnammal College for Women Coimbatore Tamilnadu India Abstract Personal identification and authentication is difficulty in all the systems. Shared secrets like Personal Identification Numbers or Passwords and key devices such as Smart Cards are not presently sufficient in few situations. These traditional tokens based systems may be easily stolen or lost. Biometrics is the only way of improving the capability to recognize the persons according to the physiological or behavioral features. In many real-world applications, unimodal biometric system suffers from some limitations of noise in sensed data, intra-class variation, inter-class similarities, non-universality and spoof attacks. Multibiometric systems seek to alleviate some of these problems by consolidating the evidence obtained from different sources. These systems help to achieve an increase in performance. This paper focused on developing a multimodal biometrics system, which uses biometrics such as iris and retina. Fusion of biometrics is performed by means of rank level fusion. The ranks of individual matchers are integrated using the borda count, and logistic regression approaches. The developed multimodal biometric system utilize and Fisher s Linear Discriminant (FLD) and Principal Component Analysis (PCA) methods for individual matchers ( and ) identification. The features from the biometrics are obtained by using the Fisherface. The experimental result shows the performance of the proposed multimodal biometrics system. 1. Introduction Biometrics means The statistical analysis of biological observations and phenomena. It refers to the use of physical (e.g., fingerprint, face, iris, retina, hand geometry) and behavioral (e.g., gait, signature, speech) characteristics for reorganization of an individual. In the biometric system, these characteristics are referred to as traits, indicators, identifiers or modalities. A simple biometric system consists of following basic modules [1]: a) Sensor module: It acquires data from a user b) extraction: It processes the data and extracts a feature set c) Matching module:it compares the extracted feature set with stored templates. d) Decision-making module: validate a claimed identity is accepted or rejected. Each biometric feature has its own strengths and weaknesses and the choice typically depends on the application. No single biometric is achieved to have effective to meet the requirements of all the applications. Multibiometrics is a relatively new approach to improve the recognition performance. is an internal organ that is well protected against damage and wears by a highly transparent and sensitive membrane. It can contain many distinctive features such as furrows, ridges and rings etc [2]. technology provides greater unique identification. is the vascular pattern of the eye. It is not easy to change and replicate the retinal vascular pattern. The patterns are different for right and left eye. That will be unique for identical twins. According to the above features iris and retina are taken to develop the proposed system. The key to successful multibiometric system is in an effective fusion scheme, which is necessary to combine the information presented by different sources. The information presented in the multimodal system can be fused at any one of the following fusion levels [3]. Sensor Level Fusion In this fusion raw data obtained from different modalities are fused. The fusion can be done using samples of same biometric trait obtained from multiple sensors or multiple instances of same biometric trait obtained using a single sensor. Level Fusion Data obtained from different sensors are first subjected to feature extraction algorithms and the feature vector, which is subsequently used for recognition. Match-Score Level Fusion 1975

2 The match score obtained from different matchers are combined together. The score normalization technique is followed to map the scores obtained from different matchers. Decision Level Fusion The final outputs obtained from multiple classifiers are combined for making the final decision. This level of fusion is also known as abstract level fusion because it is used when there is access to only decisions from individual classifiers. Rank Level Fusion It combines identification ranks obtained from multiple unimodal biometrics. In rank level fusion, three methods are used to combine the ranks assigned by different matchers such as highest rank, borda count and the logistic regression method. In sensor level fusion some noise can be present in the obtained biometric data because of improperly maintained sensors. The feature level fusion leads to dimensionality problem and also relationships between features are not known. In score level fusion an inappropriate normalization technique result leads to very low recognition performance rate. Decision level fusion contain only limited amount of information about the data. Rank level fusion is efficient technique, which is used for developing the proposed system to consolidate the multiple unimodal biometric matcher outputs. Section II describes related works of multibiometric system. Section III describes rank level fusion and its methods. Section IV describes methodology for recognition purpose. Section V will summarize the experiment results. The result shows the performance of the proposed system. 2. Related Works Kisku et al. [4] developed a multimodal biometric system using sensor level fusion scheme for face and palmprint. Rossa and Govidarajan [5] developed a multimodal biometric system for face and hand using feature level fusion with PCA (Principal Component Analysis) and LDA (Linear Discriminant Analysis) method. It improves matching performance at FAR 0.01% and GAR 80%. Y.Yao et al. [6] developed a multimodal biometric system using face and palmprint at feature level. Here, the Gabor features of face and palmprints are extracted separately. Extracted features are then analyzed using PCA. Finally the feature level fusion is carried out to form a fused feature space. Gian Luca Marcialis and Fabio Roli [7] developed for multi sensor fingerprint verification system using decision-level fusion of optical and capacitive sensors. When compared with individual sensor (the optical sensor) the proposed multisensor system improves performances. Ajay kumar and sumit shekhar [8] developed a biometric system of palmprint recognition using rank level fusion. Snelick et al. [9] proposed a multimodal system for face and fingerprint, with fusion methods at the score level. Maruf Monwar and Marina Gavrilova [10] developed a multimodal biometric system of face, ear and signature using rank level fusion with borda count and logistic regression method. We obtained EER 1.12% when compared with score fusion EER is 1.88%. M. Nageshkumar et al. [11] have proposed multimodal system for face and palmprint using match- score level fusion. The overall accuracy of the proposed system is more than 97%. Nandakumar et al. [12] proposed a minutiae and texture based fingerprint fusion study using a quality-weighted sum (QWS) rule for score level fusion. Rattani et al. [13] proposed a multibiometric system of face and fingerprint using feature level fusion. The proposed system obtained accuracy of 96.66%. Baig et al. [14] developed a multimodal system for fingerprint and iris using score level fusion, which utilizes a single hamming distance matcher. The EER of the proposed system is 2.86%. 3. Rank Level Fusion The goal of rank level fusion is to consolidate the ranks output from different biometric modalities to identify an individual. Figure 1 shows an example of multimodal biometric system of iris and retina using rank level fusion. Multimodal Biometric System Multiple Biometric Trait Enrollments Rank Result Rank Figure 1. Multimodal Biometric System Using Rank level Fusion In rank level fusion the ranks assigned are combined by different matchers. Here, the proposed system uses borda count method and logistic regression method to integrate the rank of individual matcher. Highest Rank Method In this method each possible match is assigned as a highest (minimum) rank by different matchers. Borda Count Method This method uses the sum of the ranks assigned by individual matchers to calculate the final value. Logistic Regression Method This method is a generalization of the borda count method. In this method, a weighted sum of the individual ranks is calculated. The weights are calculated during the training phase using logistic 1976

3 Identity regression method. Figure 2 shows the proposed system. Identity Rank Person 1 1 Person 2 4 Person 3 5 Person 4 3 Person 5 2 Fused score Borda Count Method Recorded rank Identity Rank 2 Person 1 Person 2 5 Person 3 3 Person 4 1 Person 5 4 Logistic regression method (=0.15, =0.5) Fused score Recorded rank Person 1 3/2= /2= Person 2 9/2= /2= Person 3 8/2= /2= Person 4 4/2= /2= Person 5 6/2= /2= Figure 2. Rank level fusion using iris and retina In the Borda count method, these ranks are added and then divided by 2 (number of biometrics) and it returns 1.5, which is the first rank. For the logistic regression method, the weights 0.15 and 0.5 are assigned for iris and retina. For Person 2, the rank is 4 and 5 from iris and retina. For the reordered rank calculation, initial ranks are multiplied by their respective weights (4 multiplied by 0.15, 5 multiplied by 0.5). Then, these two new ranks of Person 2 is added and divided by 2 (number of biometrics) and the result is 1.55, which is considered as rank 2 (Second from the lowest). 4. Methodology This section deals with the methodology used in the proposed biometrics system. Eigenimage and fisherface techniques are used for enrollment and recognition of biometric traits. Eigenimage is the first method considered as a successful technique for recognition. It uses Principal Component Analysis (PCA) to linearly project the image space to a low dimensional feature space. The fisherface method uses both PCA and LDA (also called FLD) to produce a subspace projection matrix. However, the fisherface method is able to take advantage of within-class information, minimizing variation within each class, yet still maximizing class separation. Two scatter matrices are computed A) Recognition Using Eigenimage Eigenimage feature extraction is based on the K L transform and is used to obtain the most important features from the iris, and retina. These features are obtained by projecting the original subimages into the corresponding subspaces. Create two image subspaces: one for the iris subimages and one for the retina subimages. The process of obtaining these subspaces and projecting the subimages into them is identical for all subspaces. The system is first initialized with a set of training images. Eigenvectors and eigenvalues are computed on the covariance matrix of these images. From the eigenvectors choose only a subset, which has the highest eigenvalues. The higher the eigenvalue describes the more characteristic features of an image. Eigenimages with low eigenvalues can be omitted because it contains only a small part of the characteristic features of the images. Finally, the known images are projected onto the image space, and their weights are stored. This process is repeated as necessary. After defining the eigenspace, the system projects any test image into the eigenspace. An acceptance or rejection is determined by applying a threshold. Any comparison producing a Euclidian distance below the threshold is a match. The steps for the recognition process can be summarized as follows 1. Measure the distance between the unknown image s position in the eigenspace and all the known image s positions in the eigenspace. 2. Select the image closest to the unknown image in the eigenspace as the match. In order to apply the rank-level fusion method, system needs the output of matched images which are ranked. For this, the system defines the image with the lowest distance as rank-1 image, the image with the second lowest distance as rank-2 image, and so on. Figure 3 is the representation of the proposed system. 1977

4 FAR (%) FRR (%) A Kavitha et al, Int. J. Comp. Tech. Appl., Vol 2 (6), B) Euclidian Distance One of the most popular similarity distance functions is the Euclidian distance. It is just the sum of the squared distances of two vector values (xi,yi), version of the CUHK Database and retina use DRIVE data set are used. Figure 4 shows the false rejection rate of the proposed system, from this logistic regression method provide better result than borda count method in terms of low error rate. ROC Curves for Rank Level Fusion Methods (1) C) FLD (Fisher Linear Discriminant) The fisherface method uses both PCA and LDA (also called FLD) to produce a subspace projection matrix. However, the fisherface method is able to take advantage of within-class information, minimizing variation within each class, yet still maximizing class separation. Two scatter matrices are computed.the within-class (SW), betweenclass (SB) and total (ST) distributions of the training set is, (2) (3) (4) Where the average image vector of the entire training is set and is the average of each individual class X i (person). preprocess Preprocess preprocess Preprocess Final Output Enrollment Eigen Eigen retina Eigen Eigen Rank Level Fusion Figure 3. Proposed System Figure 3. Proposed System System Database Code Code Matching Fingerprint Ranking Ranking 5. Experiment and Results The proposed system has been implemented using MATLAB. The biometrics obtained from 100 persons is used for evaluation. The performance of the proposed system is evaluated using the parameters such as False Rejection Rate (FRR) and False Acceptance Rate (FAR). For iris use Cuhk(University of hongkong). A public Persons Figure 4 ROC Curves for Ranlk Level Fusion Methods Persons Borda Count Logistic Regression Borda Count Method Logstic Regression Method Figure 5 Figure 5 shows the false acceptance rate of the proposed system.it is observed that the logistic regression method provide better result than borda count method. 6. Conclusion Biometric systems offer several advantages over traditional based methods. Multibiometrics is an efficient technique used to identify a person accurately. From the comparison results that ranklevel fusion with the logistic regression approach provided the better performance in terms of error rates at FAR of 0.08% at FRR of 85.2%. 7. References [1] Arun Ross and Anil Jain, Information Fusion in Biometrics, Appeared in Pattern Recognition Letters, Vol. 24, Issue 13, pp , September, [2] J. Daugman, "How Recognition Works", IEEE Trans. on Circuits and Systems forvideo Technology, Vol. 14, No. 1, pp , January [3] G. Hemantha Kumar and Mohammad Imran, Research Avenues in Multimodal Biometrics, IJCA Special Issue on Recent trends in Image Processing and Pattern Recognition, [4] D.R. Kisku, J. K. Singh, M. Tistarelli, and P.Gupta, Multisensor biometric evidence fusion for person authentication using wavelet decomposition and monotonic decreasing graph, in Proceedings of 7th International Conference on Adavnaces in Pattern 1978

5 Recognition (ICAPR- 2009), Kolkata, India, 2009, pp [5] A. Ross and R.Govindarajan, level fusion using hand and face biometrics, in Proceedings of SPIE Conference on Biometric Technology for Human Identification, 2004, pp [6] Y. Yao, X. Jing, and H. Wong, Face and palmprint feature level fusion for single sample biometric recognition, Nerocomputing, vol. 70, no. 7-9, pp , [7] G. L. Marcialis and F. Roli, Fingerprint verification by fusion of optical and capacitive sensors, Pattern R'ecogn. Lett, vol. 25, no. 11, pp , Aug [8] Ajay kumar and sumit shekhar, Palmprint recognition using rank level fusion, Department of Computing, The Hong Kong Polytechnic University, Hong Kong Department of Electrical and Computer Engineering, University of Maryland, College park, USA. [9] R. Snelick, U. Uludag, A. Mink, M. Indovina, and A. K. Jain, Large scale evaluation of multimodal biometric authentication using state-of the-art systems, IEEE Trans. Pattern Anal. Mach. Intell., vol. 27, no. 3, pp , Mar [10] Md. Maruf Monwar and Marina L. Gavrilova, Multimodal Biometric System Using Rank-Level Fusion Approach, IEEE Trans on systems, man, and Cybernetics, VOL. 39, no.4, [11] Nageshkumar.M, Mahesh.PK and M.N. Shanmukha Swamy, An Efficient Secure Multimodal Biometric Fusion Using Palmprint and Face Image, IJCSI International journal of computer science Issues, vol.2, [12] K. Nandakumar, A. Ross, and A. K. Jain, "Incorporating ancillary information in multibiometric systems," Handbook of Biometrics.New York: Springer-Verlag, pp , [13] A.Rattani, D.R.Kisku and M.Bisgo, level fusion of face and fingerprint biometrics, Italian Ministry of Research, the Ministry of Foreign Affairs and the Biosecure European Network of Excellance. [14] Asim Baig, Ahmed Bouridane, Fatih Kurugollu, and Gang Qu, Fingerprint Fusion based Identification System using a Single Hamming Distance Matcher, International Journal of Bio- Science and Bio- Technology, Vol. 1, No. 1, December