RESEARCH AND DEVELOPMENT IN INFORMATION AND TECHNOLOGY (SIT 740) FACULTY OF SCIENCE AND TECHNOLOGY ASSIGNMENT 01 LITERATURE REPORT ON MULTIMODAL BIOMETRIC SYSTEMS VEERKAMAL KAUR BRAR ST ID-217072824 MULTIMODAL BIOMETRIC SYSTEM 1.INTRODUCTION TO BIOMETRIC SYSTEMS Individual character alludes to an arrangement of characteristics (for example name of the person,driving license number, bank account holder and so on.) that relate to a man. Character administrator is the process of making, keeping up and devastating personalities of people in a populace(Ashish,2010). Three important approaches to verify the personality of a man are a.)something known earlier b.)something you convey c.)something really is figure A: Classification of multi-biometric system Biometric authentication system is latest technology that checks information of a person (climate identity) from physiological traits such as face, ear, iris, fingers or behavioral traits such as signature and voice. Biometric which considers only single trait as information is called unimodal biometric system but these have few drawbacks such as noisy data, none-universality, spoof attacks and frequent error Rate when the physiological over behavioral traits are combines, these from multimodal system. Already existing multimodal system are iris and ear, fingerprint and iris, many more. These it mostly depends upon unique biological traits to get the successful results in the world of IT, identification of human for security purpose has become main biometric systems. It is being used for different purpose by different organization like governmentand business organization used it for details about the populace, crime investigation agencies use it match the biometric like fingerprint and DNA, e-commerce and other organizations use it for security purpose. 1.2 How do biometric system works? In biometric system, information is recorded and compared mostly,characteristics and recorded as images, time series data is for signature recognition and a way form is used for speaker recognition. Just to increase the efficiency, characteristics are not recorded directly, it extracts features from samples and encode them in the suitable form that makes storage and comparison smoother. When the biometric system is used first time by a.) Verification phase b.) identification phase VERIFICATION PHASE- In this phase system check the identity of the person by matching it to the already saved biometrics in the system. e.g.ATM user need to fill pin code in the ATM machine to validate his/her Identity (Michael 2005). IDENTIFICATION PHASE-it is the process of positively identifying the user. In this phase, identification is made by comparing the present biometric traitseither all the individual identities in the large database (Li, yawn,2012). Enrolment (B) Identification (C) For Verification (D) 1.3 MULTIMODAL BIOMETRIC SYSTEM MOTIVATION Unimodal Biometric Authentication System is less expensive as compared to the multimodal biometric system but this system has some drawbacks such as intra-class variations, inter-class similarities, non-universal data and spoof attacks, noise issues with the data etc. To handle above mentioned drawbacks of Unimodal system, new Multimodal system were introduced. The unimodal biometric systems are not much reliable due to following reasons. They basically Work on the single biometric trait. So, this has following disadvantages (Naveen, 2011): • Non-universality: sometimes Because of weak fingerprint impressions, few people are not able to verify claimed identity i.e.their fingerprints do notmatch. • Noisy sensor data: Environmental conditions like humidity, pressure, dust Etc. are responsible for disturbance. Sometimes noisy signals are captured only. • Lack of individuality: some people have same identity due to genetic factors for example like son and father can have similar features So it is difficult to distinguish if only one biometrics has been used. • Lack of representation: The data collected for verification and testing must not match the data for training. • Susceptibility: As it is not easy to steal someone’s characteristics but sometimes spoofed characteristics like signature and voice can be used to make such criminal attacks. From the above discussion,there are high error rates so no unimodal biometric is accurate. So, these limitations can be reduced by introducing the concept of the multi-modal system (Absit, 2005). So, in biometric systems, many traits are used that means more security from an intruder to spoof all the traits simultaneously. Therefore, they are more secure, trustworthy and provide better results and hence accuracy. The concept of the multimodal systems makes it possible to use the biometric technology in all applications. FIGURE E:Fusion scenario in multimodal biometric system 2. LITERATURE REVIEW In 2015,Aboshosha, Dahshan, the authors propose that single biometric system is facing many problems, so they use multimodal biometrics to overcome the problems like corrupt data, non-integrated and lost data. Two or models are utilized in multimodal biometrics. Here authors use fusion of three individual biometric characteristics to enhance the accuracy of the system. They use the combination of fingerprints, face traits and eye i.e. iris at score level. Firstly min-max normalization is used to normalize the resulted scores. After that, to get fusion, rules like sum, product and weighted sum are used. Finally, the comparison shows that multimodal biometric system is superior to unimodal biometric system. Also, concluded that sum rule gives the best results over sum or product method. Another solution was provided Benaliouche, Touahriain2014, In this paper the authors compares the performance of three methodologies for recognition of iris and fingerprints together for multimodal biometric. He used three different approaches i.e. classical sum rule and second was weighted sum rule, last was fuzzy logic method. Min-max normalization rule is used to normalize the both resulted scores. After that score combination approach is used. The results show that fuzzy logic is the good one after the classical weighted sum rule. Parameters like matching time, fault rates and accuracy are used to evaluate the performance of these methods. Different databases are used for iris and fingerprints to conduct experiments on public that are CASIA-Iris database V1 V2 and the another EVC 2004 fingerprint database. Experimental results give priority to fusion because this logic method makes simple way and gives better results for human. In 2014, another idea was proposed by MahdaviKulkarni, problems facing by unimodal biometric systems like protection, correctness, performance and robustness. To obtain the better results, different biometric characteristics are used to gather data by the multimodal biometric system. There are three levels of fusion in multimodal approach that is sensor, feature, matching score and decision. To get the matching ranks easily matching score level is greatly preferable level as it contains a lot of data to compare. In this paper, the focus of author is on matching score level fusion as the information obtained is both feasible and practical, in multimodal biometric system. Hitesh,Wadhwa (2014)described in This work shows that there are main three categories of algorithms for advanced speech enhancement: (1) filtering/estimation based noise reduction, (2) beam forming and (3) active noise cancellation (ANC) techniques. Recently under the conditions when there is a lot of noise the GA algorithm work effectively and give better results. To choose the features that can distinguish the different words evolutionary computation is applied in the form of genetic algorithm. By doing this, the system can able to acknowledge the word-speech with real time performance by using fewer feature elements. Saishanmuga Raja (2013), uses a method including Genetic algorithm and Neural Network to identify a person on the bases of iris recognition. In this process, firstly the iris region is localized and by using the iris pattern recognition, data set of iris images created. He proposed that a neural network can be used to decrease the low identification rate, less accuracy and excessive time for recovery and the parameters of Neural Networks are processed by using genetic algorithm. A good identification rate is obtained and results also show the decrease in training time. In 2012, Ramya, Muthukumar and Kannan presented thatwhile comparing to the other features most of the time the fingerprint and the Iris are preferred as the most reliable traits. In the enrolment phase, this method consists of six steps: a) gettingtraits of fingerprint and iris b) creating Fusion of obtained features, c) a Key is produced from the fused method which value more than 128 bit which is sufficient forthe AES encryption, d) AES encryption method, e) Hash Encoding logic, f) AES decryption method. To acquire key from the fused feature hash is given to AES encryption as a message. In the next phase, decryption is to be done that is known as the verification phase. A message one gets after the encrypted value from two phases is shown to be same. Publicly available sources provide Fingerprint and CASIA Iris database provides Iris. The result also shows that there is a decrease in the false acceptance and false rejection rates. In 2012,Nguyen Viet Cuing Vu Dinh,shows that there are various activities that involves human identification. Here we are dealing with the eye movement identification problem. In the eye movement identification, the information of the movement of eye is used. A multiclass classification model is involved to overcome this problem. better results can be obtained by using Mel frequency kestrel coefficient for encoding many traits for the classification model than Fourier transform, cestrum, or raw representations to represent different eye features like position of the eye, difference of eye, and velocity of eye. Various models of classification are also compared by the authors. One of them named linear-kernel SVMs gives the good results, obtaining approximate (93.56% 91.08%) on the both large and small datasets respectively as the result of their experiment. They also performed experiments to check the contribution of each eye to the final classification accuracy. 2.2 TECHNIQUES USED IN MULTIMODAL BIOMETRIC SYSTEM Different type of fusion techniques is used in multimodal biometric system- 1) fusion prior matching 2) Fusion after matching FIGURE F: Different type of fusion levels in multi-modal biometric system FUSION PRIOR MATCHING is obtained with the help of two methods- 1.1) sensor level fusion- This type of fusion can be used only if the many sources demonstrate samples of the one biometric trait which can be achieved by using a single sensor or different sensors which are compatible. 1.2) feature level fusion-this type of fusion is obtained by joining more than one feature set which is extracted from more than one biometric source. Feature sets can be of two types.one is homogeneous and other is heterogeneous. The integration of these feature set originates problems because these feature groups are formed from different type of algorithm and modal. 2)FUSION AFTER MATCHING is obtained with the help of two methods- 2.1) match score level fusion-Matching score level fusion is the best method as compare to the others. It provides richest set of information. This fusion technique is also known by the name of measurement level or confidence level fusion. It is comparatively very easy to integrate the scores created by different biometric matchers. This method is the most commonly used method for fusion. In this method,we try to identify the pattern only in the two classes: genuine (whatsomethingis) or impostor (individual who only pretends to be someone else for example fraud). MATCH SCORE LEVEL FUSION is achieved with the help of three methods- 2.1.1) density based fusion 2.1.2) transformation based score fusion 2.1.3) classifier based score fusion As the match score level, fusion use scores fromvarious modalities which are based on different scaling methods, therefore scores cannot be combined and are used indirectly. It is essential to conduct score normalization by converting thedifferent scores into common domain or scale. Score normalization can be carried out with different methods of normalization with different modalities. 2.2) rank level fusion Rank level fusion integrates the ranks of the result by the subsystems of individual to get a perfect rank of every identity. Rank level fusion is less appropriate as compare to the former. Because it provides low information about the identity. The rank level fusion is mostlyused for the identification of a person rather than verification. In verification, we generally compare template only with other template in the database and create rank of identities in sorted form. 2.3) decision level fusion Decision level fusion is performed out at decision level only when the decisions results are made available by the individual matcher. At present, if one is behind commercial off the shelf tools for biometric verification and security, then only decision level fusion is the only option for fusion, as they don’t give any information about the scores or features.Moreover,they do not provide information about the ranking of different users after comparison. Decision level fusion is also known by abstract level fusion. They only identify whether the user is genuine or imposter after matching. With decision level fusion, there are many logic and rulesthat can be used to check and verify the user. The most frequently used approach for decision level fusion is majority voting. In this identity is provided to the input sample for which the large number of matchers are checked. AND and OR rules are used less, because in this we combine two different matchers, so there are chances that there might be degradation of performance. 3.PROBLEM FORMULATION It a lot of previous work and research has already been done in the field of combining different biometric techniques and creating different fusion of the biometrics. Each one used different set of rules and techniques. Although, there are various fusion techniques. each has its own benefits and drawbacks. Proposed idea is to design a new technique to overcome drawbacks, get better performance and results. the biggest trouble with any fusion technique is that matching and identify features after the extraction for better alignment of the feature value. There are a lot of algorithms and techniques which are available for the feature extraction but there is no unique method which can align them all together at one place. better optimization techniques could have been used for better results. 4. CONCLUSION In today’s environment, commercial multimodal biometric systems are becoming very popular and they resolve many problems present in unit-modal biometric systems. Theyprovide better identification performance,escalated population coverage, data spoofing, and improves indexing. By combining two or more sources of information,these systems have increased matching performance. The system would take the security to another level. As explained above in the different research papers, a lot of work has been done by the researchers on the multimodal biometric systems. Each one has used different set of rules and techniques. These rules provide different fusion methods and helps in the correct identification of a person. 5.REFERENCES • Jaina A, Nandakumara K, Rossb G,2005,” Scorenormalizationinmultimodalbiometricsystems” • Aboshosha A,Kamal A, Eman A.,,2008,"Score level Fusion For Fingerprints, Iris andFace Biometrics" • Kulkarni M,2014,"Study of Multimodal Biometric System: A Score Level Fusion Approach.” • Jaina,NandakumaraK,RossbA,2015,”Scorenormalizationinmultimodalbiometricsystems” • www.google.com • www.googlescholar.com • P. D. Garje, Agrawal s, 2012, “Multibiometric Identification System On Score Level Fusion” Ramya n, Muthukumar a and Kannan s, 2012, “Multibiometric based • authentication using feature level “ • Boodoo b, Subramanian k, 2009, “Robust Multi-Biometric Recognition Using Face and Ear Images” • Ross and K. Jain,2005,” MULTIMODAL BIOMETRICS: AN OVERVIEW” • www.iwsinc.com • www.igi-global.com