International Journal of Engineering Technology and Computer Research (IJETCR) Available Online at www.ijetcr.org
Volume 5; Issue 5; September-October: 2017; Page No. 80-91 Journal Approved by UGC
Multimodal Biometric Authentication Technique with Optimized Feature Selection Using Crow Search Optimization Algorithm
T. Srinivasa Rao1, Dr. E. Srinivasa Reddy2
1,2Computer Science & Engineering, Acharya Nagarjuna University, Guntur, India,
1[email protected], 2[email protected]
Received 14 Aug. 2017; Accepted 14 Sep. 2017
Abstract
Biometric recognition is a process of recognizing an individual with their physiological and behavioral biometric traits. In this paper, a multimodal biometric system is proposed by combining the scores of fingerprint, palmprint and speech traits of a person. This information fusion takes place at the matching score level. Score normalization is a technique to transform the obtained scores into a uniform domain, prior to combining them.
The resulting scores are compared to a threshold value for taking a decision of accepting or rejecting the person.
The recognition accuracy of fusion methods strongly depends upon the correctness of this threshold value.
Hence, we propose crow search optimization (CSO) technique for selecting the optimal threshold value for each of the fusion method employed. The experimental results obtained using finger print, speech and palm print databases show that the application of CSO results in higher recognition rates and lower error rates. To the best of our knowledge, it is the first work that applies CSO to enhance the accuracy of biometric authentication process.
Key Words: multi-modal biometrics, biometric fusion, fingerprint, palmprint, speech, score level fusion.
INTRODUCTION
Biometrics are the representations of physical and behavioral traits of humans for verifying or determining the identity of humans [1]. Biometric recognition is a process to recognize an individual using physiological and behavioral biometrics traits.
The physiological characteristics such as face, iris, fingerprint, palm and the behavioral characteristics such as voice, gait, key stroking and signature.
Biometrics are considered more reliable than traditional passwords or tokens, as these traits are permanently associated with the human. Biometrics offers some more advantages than the traditional security measures such as non-repudiation, accuracy and security. Number of biometric authentication systems have been proposed based on various biometrics. Recently, several studies reveal that the walking pattern of a person can also be used for recognition purpose.
A single biometric trait is used for authentication system which is called the unimodal system [2].
Unimodal biometric systems have many drawbacks such as noisy data, intra-class variation, non- universality and spoof attacks. Some drawbacks of
the unimodal system can be overcome using multiple biometrics instead of a single biometric in the recognition process. Various biometric fusion prove that they increase accuracy and decrease the vulnerability to spoofing. Such systems are called multimodal biometric systems. Multimodal biometric systems are much reliable because features of the various biometrics of a particular person are used and more information of a person is used for recognition.
In multimodal biometric system, information fusion can be done at four levels [3, 4, 5]. They are 1. fusion at the sensor level, 2. Fusion at feature extraction level, 3. Fusion at matching score level and 4. Fusion at the decision level. Sensor level fusion is the process of combing data of two or more biometric traits from the biometric sensors. Feature-level fusion is the process of combining feature vectors obtained either by different sensors or by applying different feature extraction algorithms on the same data.
Score level fusion is the process of combining different matching scores obtained by different biometric traits. Decision level fusion is the process of combining the decisions taken by the different biometric systems. The process of matching score level fusion has the following two steps: 1.
normalization and 2. fusion. The normalization is the process of converting the scores of different traits to a comparable range of values. If we do not perform
normalization, a biometric with a higher range could eliminate the contribution of another with a lower one.
Figure 1: various levels of fusion in multimodal biometrics system, (a) sensor-level fusion, (b) feature level fusion, (c) matching score level fusion and (d) decision-level fusion
2. PROPOSED METHOD
In this paper, a multimodal biometric system is proposed by combining the scores of fingerprint, palmprint and speech traits of a person. This information fusion takes place at the matching score level. Score normalization task is performed to transform the scores into a common domain, before combining them. The resulting scores are compared to a threshold value for taking a decision of accepting or rejecting the person. The recognition accuracy of fusion methods strongly depends upon the correctness of this threshold value. Hence, we propose crow search optimization (CSO) [4]
technique for selecting the optimal threshold value for each of the fusion method employed. As the
information was taking from different sources, it is needed to keep the template secure from the database. So, to secure the template, RSA public key cryptographic algorithm is used to encrypt the template [5] in the proposed system. To overcome the problems faced by individual traits of palmprint, fingerprint and speech, a combination is proposed for the recognition system. The proposed system also provides anti spoofing mechanism by making it very difficult for an intruder to spoof multiple biometric traits simultaneously. The scores obtained from individual biometric traits are combined at matching score level [9, 10] using weighted sum of score technique.
Figure 2: Architecture of the proposed method
2.1 FINGERPRINT FEATURE EXTRACTION
Fingerprint which contains the impressions with the distinct ridges on the finger tips can be extracted by using the Finger print recognition technology [6].
There are some interesting points called minutiae which are by the corners or forking of the abrasion skin ridges on every finger. In this paper, Minutiae extraction is carried out based on bifurcation i.e. with the point at which a single ridge splits into two ridges and termination i.e. immediate ending of a ridge.
Fig.3 (a) Original image
(b) Extracted minutiae points
2.2 PALMPRINT FEATURE EXTRACTION BY TEXTURE ANALYSIS
The feature extraction method for palm-print has two steps: filtering and matching. We will discuss about the Gabor filter first which is the motivation for our palm print research.
A) Gabor function
In this paper, we tried to apply the Gabor filter for the authentication of palm print based on all these properties of Gabor functions. The general form of a circular 2-D Gabor filter in the spatial domain is
𝐺𝐺(𝑥𝑥, 𝑦𝑦, 𝜃𝜃, 𝑢𝑢, 𝜎𝜎) = 1
2𝜋𝜋𝜎𝜎2𝑒𝑒𝑥𝑥𝑒𝑒 �−𝑥𝑥2+ 𝑦𝑦2
2𝜎𝜎2 � × 𝑒𝑒𝑥𝑥𝑒𝑒{2𝜋𝜋𝜋𝜋(𝑢𝑢𝑥𝑥 cos 𝜃𝜃 + 𝑢𝑢𝑦𝑦 sin 𝜃𝜃)}
where i =√−1; 𝜃𝜃 controls the orientation of the function; σ is the standard deviation of the Gaussian
envelope and 𝑢𝑢 is the frequency of the sinusoidal wave.
B) Filtering and feature extraction
A Gabor function defined as G(x, y, Ɵ, u, σ) with a the set of parameters (σ, Ɵ, u), is changed into a discrete Gabor filter, G[x, y, Ɵ, u, σ]. To achieve good robustness to brightness, the Gabor filter is converted to zero DC (direct current) by use of the following formula:
𝐺𝐺�[𝑥𝑥, 𝑦𝑦, 𝜃𝜃, 𝑢𝑢, 𝜎𝜎] = 𝐺𝐺[𝑥𝑥, 𝑦𝑦, 𝜃𝜃, 𝑢𝑢, 𝜎𝜎] −∑𝑛𝑛𝜋𝜋=−𝑛𝑛∑𝑛𝑛𝑗𝑗=−𝑛𝑛𝐺𝐺[𝜋𝜋, 𝑗𝑗, 𝜃𝜃, 𝑢𝑢, 𝜎𝜎]
(2𝑛𝑛 + 1)2
where (2n+1)2 is the size of the filter. As the Gabor filter has odd symmetry, automatically the illusionary part of the Gabor filter contains zero DC.
2.3 SPEECH FEATURE EXTRACTION USING MEL FREQUENCY CEPSTRUM COEFFICIENTS (MFCC)
MFCC is the most famous and reliable feature extraction technique for speech recognition. The frequency bands are stored as logarithmic values in MFCC. So that it can estimate the human system response more relatively than any other system. The frequency bands in Mel - frequency cepstrum are spread evenly on the Mel scale and MFCC can be obtained from Mel - frequency cepstrum. MFCC vector can be calculated from each frame and this is based on the short-term analysis. The following formula is used to calculate MFCC
Mel (f) =2595*log10 (1+f/700)
The steps involved in MFCC feature extraction are as shown in the following figure.
Figure 4: MFCC Feature extraction stages 2.4 CROW SEARCH ALGORITHM
The Crow Search Optimization algorithm [4] is a meta-heuristic algorithm proposed by Askarzadeh in 2016. The key inspiration of this algorithm came from crow search mechanism for hiding their food. Crows are considered as one of the most intelligent birds.
Crows a have large brain when compared to their body size. Crows are clever. Their brain is slightly lower in the brain of humans. Furthermore, they are self-aware in mirror test. They are able to remember faces. If they find an unfriendly one, they warn the other crows in a special way of communication.
Further, they can store and remember their food for several months. They are also known to be thieves, as they steal the food of the other birds. They make use of their experience as a thief in predicting pilferer’s behavior. They are very cautious. Whenever a crow reports a thievery, other crows move their hiding places. This kind of behavior enables them to avoid any kind of future victims.
2.4.1 MATHEMATICAL MODEL OF CSA
Consider the number of crows denotes as M. D is the total number of dimensions, yj,t is the position of j crow at iteration t in the search space, where j = 1,2,...M. tMax is the maximum number of iterations.
Every crow has memory of place of hiding. Nj,t is the position of hiding place at iteration t for crow j. Nj,t is defined as the best position found so far by crow j. At an iteration t, crow j wants to follow the hiding place of crow z. In this case, there are two possible cases that can happen:
Case 1: Crow z don’t not know that crow j follows it, Crow j follows to the hiding place of crow z. The position of crow j is updated as follows:
𝑦𝑦𝑗𝑗 ,𝑡𝑡+1 = 𝑦𝑦𝑗𝑗 ,𝑡𝑡+1+ 𝑅𝑅𝑗𝑗 × 𝑓𝑓𝑓𝑓𝑗𝑗 ,𝑡𝑡× �𝑁𝑁𝑧𝑧,𝑡𝑡− 𝑦𝑦𝑗𝑗 ,𝑡𝑡�
where fl denotes the flight length. Rj is a random number ∈[0, 1]. fl has a great influence on searching capability. fl with small value leads to local search, while the large value leads to global search.
Case 2: Crow z knows that crow j follows it and Crow z will change its position in the search space to protect its cache.
The previous two cases are mathematically defined as follows
𝑦𝑦𝑗𝑗,𝑡𝑡+1= �𝑦𝑦𝑗𝑗,𝑡𝑡+1+ 𝑅𝑅𝑗𝑗 × 𝑓𝑓𝑓𝑓𝑗𝑗,𝑡𝑡× �𝑁𝑁𝑧𝑧,𝑡𝑡− 𝑦𝑦𝑗𝑗,𝑡𝑡�, 𝑅𝑅𝑧𝑧≥ 𝑞𝑞𝑞𝑞𝑃𝑃𝑗𝑗,𝑡𝑡 𝐶𝐶ℎ𝑜𝑜𝑜𝑜𝑜𝑜𝑒𝑒 𝑎𝑎 𝑟𝑟𝑎𝑎𝑛𝑛𝑟𝑟𝑜𝑜𝑟𝑟 𝑒𝑒𝑜𝑜𝑜𝑜𝜋𝜋𝑡𝑡𝜋𝜋𝑜𝑜𝑛𝑛, 𝑂𝑂𝑡𝑡ℎ𝑒𝑒𝑟𝑟𝑒𝑒𝜋𝜋𝑜𝑜𝑒𝑒
where Rz is a random number ∈[0, 1] and AP is the awareness probability of crow z at t iteration, which controls the balance between exploration and exploitation. Small values of AP lead to the search on local regions (exploitation), while the large value leads to global search in the search space (exploration). Crow Search Algorithm starts with setting the constraints, D, tMax, M, AP, and fl. Every
crow’s position y is randomly initialized at search space. In the beginning, the crow y does not have experience to hide their food. Thus, they hide their food at initial positions N.
As the algorithm runs, every crow is evaluated using a predefined fitness function. Then, according to the fitness value, the crows update their positions using below equation. Every new position is checked for feasibility. The crows update their memory as follows:
𝑁𝑁𝑗𝑗 ,𝑡𝑡+1= �𝑦𝑦𝑗𝑗,𝑡𝑡+1, 𝐹𝐹𝑛𝑛(𝑦𝑦𝑗𝑗,𝑡𝑡+1)𝜋𝜋𝑜𝑜 𝑏𝑏𝑒𝑒𝑡𝑡𝑡𝑡𝑒𝑒𝑟𝑟 𝑡𝑡ℎ𝑎𝑎𝑛𝑛 𝐹𝐹𝑛𝑛(𝑁𝑁𝑗𝑗 ,𝑡𝑡)
𝑁𝑁𝑗𝑗,𝑡𝑡, 𝑂𝑂𝑡𝑡ℎ𝑒𝑒𝑟𝑟𝑒𝑒𝜋𝜋𝑜𝑜𝑒𝑒
where Fn() is the objective function. When the termination criteria are met, the best position is reported as the optimal solution. The pseudo code of CSA is defined in Algorithm 1.
2.5 MODIFIED DRSA (MDRSA)
The multimodal biometric system means individual information was extracted from the multiple traits.
As the information was taking from different sources, it is needed to keep the template secure from the database. So, to secure the template, RSA public key cryptographic algorithm is used to encrypt the template in the proposed system. When compared with the prevailing RSA [6] method, to achieve acceptable reputation of the decrypted image, by using the symmetry properties of the algorithm, the adjustments had done in the decryption stage of RSA.
2.6 FUSION AND DECISION
Every biometric matcher provides a similarity score indicating the proximity of the input feature vector with the template feature vector. These scores can be combined to declare the accuracy of the claimed
identity. Techniques such as sum score fusion may be used to combine the matching scores reported by the multiple marchers. The following figure 5 shows the Fusion at the matching score Level.
Figure 5: Score level fusion technique In this paper, we developed the sum score fusion by
using Face and Finger print modalities where the finger print match score is calculated by using minutiae matching, palmprint using MFCC and speech by using Gabor filter. The figure 5 shows the experimental set up for the sum score level fusion.
The results generated from the individual traits are good but the problem arises for fingerprint recognition system is the presence of scars and cuts.
The scars of the fingers add noise to the fingerprint image which cannot be enhanced fully using enhancement module. Thus, the system takes noisy fingerprint as input which is not able to extract the minutiae points correctly and in turn, leads to false recognition of an individual. Thus, to overcome the problems faced by individual traits of palmprint, fingerprint and speech, a combination is proposed for the recognition system. The fusion system provides anti spoofing measures also by making it difficult for an intruder to spoof multiple biometric traits simultaneously. Scores obtained from each and every single traits are combined at matching score level using weighted sum of score technique. Let 𝑀𝑀𝑀𝑀𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓𝑒𝑒𝑟𝑟 ,
𝑀𝑀𝑀𝑀𝑒𝑒𝑎𝑎𝑓𝑓𝑟𝑟 and 𝑀𝑀𝑀𝑀𝑜𝑜𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ be the matching scores
obtained from fingerprint, palmprint and speech modalities respectively.
1) Score Normalization: The scores are normalized and values are brought in between 0 and 1. The normalization of three of the scores are done by 𝑁𝑁𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓𝑒𝑒𝑟𝑟 = 𝑀𝑀𝑀𝑀𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓𝑒𝑒𝑟𝑟 − 𝑟𝑟𝜋𝜋𝑛𝑛𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓𝑒𝑒𝑟𝑟
𝑟𝑟𝑎𝑎𝑥𝑥𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓𝑒𝑒𝑟𝑟 − 𝑟𝑟𝜋𝜋𝑛𝑛𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓 𝑒𝑒𝑟𝑟
𝑁𝑁𝑒𝑒𝑎𝑎𝑓𝑓𝑟𝑟 = 𝑀𝑀𝑀𝑀𝑒𝑒𝑎𝑎𝑓𝑓𝑟𝑟 − 𝑟𝑟𝜋𝜋𝑛𝑛𝑒𝑒𝑎𝑎𝑓𝑓𝑟𝑟 𝑟𝑟𝑎𝑎𝑥𝑥𝑒𝑒𝑎𝑎𝑓𝑓𝑟𝑟 − 𝑟𝑟𝜋𝜋𝑛𝑛𝑒𝑒𝑎𝑎𝑓𝑓𝑟𝑟 𝑁𝑁𝑜𝑜𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ = 𝑀𝑀𝑀𝑀𝑜𝑜𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ − 𝑟𝑟𝜋𝜋𝑛𝑛𝑜𝑜𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ
𝑟𝑟𝑎𝑎𝑥𝑥𝑜𝑜𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ − 𝑟𝑟𝜋𝜋𝑛𝑛𝑜𝑜𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ
Where 𝑟𝑟𝜋𝜋𝑛𝑛𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓𝑒𝑒𝑟𝑟 and 𝑟𝑟𝑎𝑎𝑥𝑥𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓𝑒𝑒𝑟𝑟 are the minimum and maximum scores for fingerprint trait by using minutiae matching. 𝑟𝑟𝜋𝜋𝑛𝑛𝑒𝑒𝑎𝑎𝑓𝑓𝑟𝑟 and 𝑟𝑟𝑎𝑎𝑥𝑥𝑒𝑒𝑎𝑎𝑓𝑓𝑟𝑟 are the minimum and maximum scores for palmprint trait by using Gabor filter and 𝑟𝑟𝜋𝜋𝑛𝑛𝑜𝑜𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ and 𝑟𝑟𝑎𝑎𝑥𝑥𝑜𝑜𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ are the minimum and maximum scores for speech trait by using MFCC.
2) Fusion: The three normalized similarity scores 𝑁𝑁𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓𝑒𝑒𝑟𝑟, 𝑁𝑁𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ and 𝑁𝑁𝑃𝑃𝑎𝑎𝑓𝑓𝑟𝑟 are fused linearly using sum rule as
𝑀𝑀𝑀𝑀 = 𝛼𝛼 × 𝑁𝑁𝑓𝑓𝜋𝜋𝑛𝑛𝑓𝑓𝑒𝑒𝑟𝑟 + 𝛽𝛽 × 𝑁𝑁𝑃𝑃𝑎𝑎𝑓𝑓𝑟𝑟 + 𝛿𝛿 × 𝑁𝑁𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠 ℎ
Where 𝛼𝛼 , 𝛽𝛽 and 𝛿𝛿 are three weight values that can be determined using some function. In this paper, a combination of linear an exponential function is used.
The value of weight is assigned linearly if the value of matching score is less than the threshold; otherwise exponential weight age is given to the score. The value of MS is used as the matching score. So, if MS is found to be more than the given threshold value is accepted otherwise it is rejected.
3. PERFORMANCE CRITERIA
The verification system is measured using false rejection rate (FRR) and false acceptance rate (FAR) as defined in the following equations:
False Rejection Rate (FRRi): is an average of number of falsely rejected transactions. If n is a transaction and x(n) is the verification result where 1 is falsely rejected and 0 is accepted and N is the total number of transactions then the personal False Rejection Rate for user i is
𝐹𝐹𝑅𝑅𝑅𝑅𝜋𝜋 = 1
𝑁𝑁� 𝑥𝑥(𝑛𝑛)
𝑁𝑁
False Acceptance Rate (FAR𝜋𝜋=1 i) is an average of number of falsely accepted transactions. If n is a transaction and x(n) is the verification result where 1 is a falsely
accepted transaction and 0 is genuinely accepted transaction and N is the total number of transactions then the personal False Acceptance Rate for user i is 𝐹𝐹𝑞𝑞𝑅𝑅𝜋𝜋 = 1
𝑁𝑁� 𝑥𝑥(𝑛𝑛)
𝑁𝑁
Genuine Acceptance Rate (GAR): This is defined as a 𝜋𝜋=1
percentage of genuine users accepted by the system.
It is given by GAR=100-FRR.
4. RESULTS AND DISCUSSION
The GUI in MATLAB is used to design the security level of the proposed multimodal biometric system.
At first, three biometric traits fingerprint, palm print and speech are considered for multimodal fusion. The feature extractions of fingerprint, palm print and speech are done using different techniques like minutia extraction, Gabor feature extraction and MFCC method respectively and then optimized features are extracted using crow search optimization algorithm and the obtained features are encrypted using RSA algorithm for template security and stored in database and then matching score is obtained using query image, the matching scores are then fused using score level fusion. The steps involving in this process are shown in Fig. 6.
Figure 6: The complete process involved in proposed multimodal biometric system
The simulation model is developed after each biometric trait such as fingerprint, palm print and speech are trained with unimodal identity. For both unimodal and multimodal biometric systems, and using RSA and without using RSA, False Acceptance Rate (FAR) and Genuine Acceptance Rate (GAR) are computed depending upon the matching performance. The system has to yield keys for query user when the verification is essential for the system. This was shown in Fig. 7.
Figure 7: Key generation process
Figure 8: Decryption of Fused Biometric from database
The generated key is used to decrypt the template after it was searched in the database for matching. The decrypted template is used to match with current query where the current query is a fusion of three biometric traits fingerprint, palm print and speech based on fused matrix values using correlation as in Fig.8 and Fig.9.
Figure 9: system showing successful authentication
Figure 10: System showing Access Denied
Figure 11: GAR without CSA
Figure 12: GAR with CSA
When the query template is compared with the stored template, if they are not matched, system reject the request of user to access the database as shown in Figure 10. To analyze the performance of identifying an individual’s authorization of the proposed system, the performance of the enrollment module and that of the verification module for the current query template are compared. False acceptance rate and genuine acceptance rate are computed depending upon the genuine and fraud authentication during a verification module. The false acceptance rate defines about the fraud user allowance in authentication and it should be low.
Whereas genuine acceptance rate defines about the genuine user allowance and it should be high. The performance of multimodal biometric system was compared with unimodal and then a plot was drawn on FAR versus GAR.
The samples of unimodal biometric traits such as fingerprint, palm print and speech are trained without CSA and then multimodal biometric trait
such as fingerprint, palm print and speech was also implemented using score level fusion techniques without CSA. The fingerprint biometric was trained and performance was calculated using FAR and GAR based on matching minutiae points with current query image for an identity. GAR of 94.32% and FAR of 5.68% for fingerprint. Similarly, the palm print was trained and tested and its GAR of 89.12% and FAR of 10.88%. The speech was trained and verified, its GAR of 91.56% and FAR of 8.44%. The multimodal biometric was trained based on fused matrix values using correlation and its GAR of 94.62% and FAR of 6.38%. Figure 11 depicts the GAR performance without RSA in the graphical format and figure 13 depicts FAR performance without RSA.
The samples of unimodal biometric traits such as fingerprint, palm print and speech are trained with CSA and then multimodal biometric trait such as fingerprint, palm print and speech was also implemented using score level fusion techniques with optimized features selected by CSA. The fingerprint biometric was trained and performance was calculated using FAR and GAR based on matching minutiae points with current query image for an identity. GAR of 95.12% and FAR of 4.88% for fingerprint. Similarly, the palm print was trained and tested and its GAR of 90.58% and FAR of 9.42%. The speech was trained and tested, its GAR of 92.45% and FAR of 7.55%. The multimodal biometric was trained based on fused matching scores using correlation and its GAR of 99.5% and FAR of 0.5%. Figure 12 depicts the GAR performance with CSA in the graphical format and figure 14 depicts the FAR performance with CSA.
Figure 13: FAR without CSA
Figure 14: FAR with CSA
We can clearly observe from figures 11, 12, 13 and 14, there is a clear improvement in the GAR using optimized features of CSA than GAR without CSA.
There is a reduction in FAR using CSA than FAR without CSA. The performance of multimodal biometric (fusion of fingerprint, palm print and speech) based on fused scores using optimized features of CSA has a GAR of 99.5% and FAR of 0.5%
whereas without CSA, GAR was 94.62% and FAR was 6.38%. The comparative curves were shown in Fig.14.
However, in order to increase GAR accuracy of multimodal biometric system as a whole, fusion at matching score level, and encryption using RSA security algorithm has been performed. Table 1 below shows the summary of results obtained during experimentation.
Table 1: Obtained GAR and FAR scores with and without CSA
Biometric Model Without CSA With CSA
GAR FAR GAR FAR
FingerPrint 94.32 5.68 95.12 4.88
PalmPrint 89.12 10.88 90.58 9.42
Speech 91.56 8.44 92.45 7.55
Fused Finger+Palm+Speech 94.62 6.38 99.5 0.5
The overall accuracy of multi-modal system has reduced FAR of 3.12% and increases GAR of 5.12%, respectively, and its performance was compared to unimodal biometric systems such as fingerprint, palmprint and speech with RSA. The performance of multimodal biometric system based on fused matching scores using optimized features of CSA have GAR of 99.5% and FAR of 0.5%, CSA with fingerprint have GAR of 95.12% and FAR of 4.88%, RSA with palm print have GAR of 90.58% and FAR of 9.42%, RSA with speech have GAR of 92.45% and FAR of 7.55%. Figure 15 and 16 clearly depicts the GAR versus FAR without and with CSA applied. It is clear that multimodal biometric traits such as fingerprint, palmprint and speech using optimized features of CSA, has increased the GAR performance and reduced the
FAR. Figure 15: GAR vs FAR without RSA
Figure 16: GAR vs FAR with RSA 5. CONCLUSIONS
This paper develops a Multimodal biometric authentication system using a combination of fingerprint, palmprint and speech where the optimized features are selected using crow search algorithm. The unimodal biometric recognition system has drawbacks due to different noisy data, non-universality of biometric data, and susceptibility of spoofing. The multimodal biometric system improves the performance of the system. In this paper, we have developed a method to select different optimized features selected by CSA from Fingerprint, palm print and speech which improves the accuracy rate than single biometric based system.
The experimental results depict that the accuracy of the proposed system would increase on combining the traits. The system is giving an overall accuracy of 99. 5 % with FAR of 0.5%.
REFERENCES
1. Nirmala Saini, Aloka Sinha, "Face and palmprint multimodal biometric systems using Gabor–
Wigner transform as feature extraction", Pattern Analysis Applications, Vol.18, pp 921–932, 2015.
2. Mohsen Tabejamaat, "Selective Algorithm Outline (SAO); An Alternative Approach for Fusing Different Palm-Print Recognition Algorithms", Neural Process Letters, vol 43, pp 709–726. 2016.
3. Nadia Nedjah1·Rafae, Soares Wyant, Luiza de Macedo Mourelle, "Efficient biometric palm-print matching on smart-cards for high security and privacy", Multimed Tools Applications, December 2016.
4. Gehad Ismail Sayed, Aboul Ella Hassanien, Ahmad Taher Azar, "Feature selection via a novel chaotic crow search algorithm", Neural Computing Applications, DOI 10.1007/s00521-017-2988-6, April 2017.
5. Samik Chakraborty, Madhuchhanda Mitra, Saurabh Pal, "Biometric analysis using fused feature set from side face texture and electrocardiogram", IET Science, Measurement &
Technology, Vol.11 Issue.2, pp. 226-233, 2017.
6. Edlira Martiri, Marta Gomez-Barrero, Bian Yang, Christoph Busch, "Biometric template protection based on Bloom filters and honey templates", IET Biometrics, Vol. 6 Iss. 1, pp. 19-26, 2017.
7. Panagiotis Moutafis, Ioannis A. Kakadiaris, "Can We Do Better in Unimodal Biometric Systems? A Rank-Based Score Normalization Framework", IEEE Transactions On Cybernetics, Vol. 45, No. 12, pp 2654-2668, December 2015.
8. Yunlian Sun, Kamal Nasrollahi, Zhenan Sun, Tieniu Tan, "Complementary Cohort Strategy for Multimodal Face Pair Matching", IEEE Transactions On Information Forensics And Security, Vol. 11, No. 5, pp 937-951, May 2016.
9. Mohammad Haghighat, Mohamed Abdel- Mottaleb, Wadee Alhalabi, "Discriminant Correlation Analysis: Real-Time Feature Level Fusion for Multimodal Biometric Recognition", IEEE Transactions On Information Forensics And Security, Vol. 11, No. 9, pp 1984-1997, September 2016.
10. L. Mezai and F. Hachouf, "Score-Level Fusion of Face and Voice Using Particle Swarm Optimization and Belief Functions", IEEE Transactions On Human-machine Systems, Vol.
45, No. 6, pp 761-773, December 2015.
11. Kien Nguyen, Simon Denman, Sridha Sridharan, Clinton Fookes,"Score-Level Multibiometric Fusion Based on Dempster–Shafer Theory Incorporating Uncertainty Factors", Ieee Transactions On Human-machine Systems, Vol.
45, No. 1, pp 132-141, February 2015.
12. Muhtahir O. Oloyede, Gerhard P. Hancke,
"Unimodal and Multimodal Biometric Sensing Systems: A Review", IEEE Access, pp 7532-7555, November 2016.
13. K. G. Srinivasa, Soumya Gosukonda, "Continuous multimodal user authentication: coupling hardand soft biometrics with support vector machines to attenuate noise", CSIT, CSI Publications, pp 129-140, September 2014.
14. Hataichanok Saevanee, Nathan L. Clarke, and Steven M. Furnell,"Multi-modal Behavioural Biometric Authentication for Mobile Devices", IFIP International Federation for Information Processing, pp 465-474, 2012.
15. Manjunathswamy B E, Appaji M Abhishek, Thriveni J, Venugopal K R, L M Patnaik,
"Multimodal Biometric Authentication using ECG and Fingerprint", International Journal of Computer Applications, Volume 111 – No 13, pp 33-40, February 2015.
16. M. Indovina, U. Uludag, R. Snelick, A. Mink, A.
Jain, "Multimodal Biometric Authentication Methods: A COTS Approach", International Journal of Computer Science and Network Security, pp 1-7, March 2009.
17. Gunjan Jiwnani, Nisheeth Saxena, "Multi-modal Biometric Authentication using Fingerprint and Iris: a Review", International Journal of Computer Science & communication Networks, Vol 5(2), pp 115-119, 2014.
18. A. Annis Fathima, S.Vasuhi, N.T.Naresh Babu, V.Vaidehi, Teena Mary Treesa,"Fusion Framework for Multimodal Biometric Person Authentication System", International Journal of Computer Science, February 2014.
19. Mohamed Soltane and Mimen Bakhti, "Multi- Modal Biometric Authentications: Concept Issues
and Applications Strategies", International Journal of Advanced Science and Technology Vol.
48, pp 23-60, November, 2012.
20. D. Jagadiswary, D. Saraswadya, "Biometric Authentication using Fused Multimodal Biometric", International Conference on Computational Modeling and Security (CMS 2016), Procedia Computer Science 85, pp 109 – 116, 2016.
21. Nidhi Desai, Kinnal Dhameliya, Vijayendra Desai,
"Feature Extraction and Classification Techniques for Speech Recognition: A Review", International Journal of Emerging Technology and Advanced Engineering, Volume 3, Issue 12, pp 367-371, December 2013.
22. Namrata Dave, "Feature Extraction Methods LPC, PLP and MFCC In Speech Recognition", International Journal For Advance Research In Engineering And Technology, Volume 1, Issue VI, pp 1-5, July 2013.
23. Shreya Narang, Ms. Divya Gupta,"Speech Feature Extraction Techniques: A Review", International Journal of Computer Science and Mobile Computing, Vol.4 Issue.3, pp. 107-114, March 2015.
24. Wai Kin Kong, David Zhang, Wenxin Li, "Palmprint feature extraction using 2-D Gabor filters", Pattern Recognition, vol 36, pp 2339–2347, 2003.
Authors Profile
T. Srinivasa Rao, Professor & TPO in Vasireddy Venkatadri Institute of Technology and Research scholar of Acharya Nagarjuna University, Guntur. He did his Master of Technology in Computer Science Engineering. He also did Mater of Science (Tech)in Mathematics from JNTU, Hyderabad. His research areas of interest are in various domains including Digital Image Processing.
Dr. E. Srinivasa Reddy, Professor & Dean R&D in University Engineering College of Acharya Nagarjuna University. He did Master of Technology in Computer Science Engineering from sir Mokhsagundam Visweswariah University, Bangalore. He also did Mater of Science from Birla Institute of Technological Sciences, Pilani. He did Philosophical Doctorate in Computer Science Engineering from Acharya Nagarjuna University, Guntur. Prof.E.S.Reddy, guided successfully more than twenty research scholars for PhD degree. His research areas of interest are in various including Digital Image Processing
