Some of the commonly challenges encountered by biometric systems are listed below:
a) Noise in sensed data: Noisy data may result from defective or improperly maintained sensors or unfavourable ambient conditions or by human beings variations in their biometric traits. Noisy biometric data may not be successfully matched with corresponding templates in the database, result a genuine user being incorrectly rejected.
b) Intra-class variations: It is caused by an individual who is incorrectly interacting with the sensor or due to change in biometric characteristic over a period of time. Although it can be handled by updated these template over time but can’t be fully remove this limitation.
c) Inter-class similarities: Overlap of feature spaces corresponding to multiple classes or individual.
d) Distinctiveness: When a biometric trait is expected to vary significantly across individuals, there may be large inter-class similarities in the feature sets used to represent these traits.
e) Non universality: Whenever every user is expected to possess the biometric trait being acquired, in reality it is possible for a subset of the users to not possess a particular biometric.
f) Spoof attacks: An impostor may attempt to spoof the biometric trait of a legitimate enrolled user in order to circumvent the system and this attack is especially relevant when behavioural traits such as voice and signature are used. However, physiological traits are also susceptible to spoof attacks .
1.2 Why multimodal biometric system?
Over the limitations of unimodal biometric system multimodal biometric system provide some advantages and more enhance security and acceptability.
a) Accuracy: Multimodal biometric acquires information from two or more biometrics trait – (e.g. fingerprint and face; or face and iris) where as unimodal biometric systems acquire information from one biometric trait– (e.g. fingerprint, face, iris, signature, voice, hand geometry, etc.). The accuracy of multimodal biometric system is basically calculated in terms of image acquisition errors and matching errors. Image acquisition errors consist of failure-to-enroll (FTE) rate and failure-to-acquire (FTA) rate whereas comprise FNMR (false non-match) rates in which a legitimate user is rejected and a (FMR) false match rate in which an impostor is authorized to access. Multimodal biometric systems have almost zero FMR, FTE & FTA rates because in this system, each and every subsystem has a determination on the person’s claim. The decision module uses different fusion strategies to combine each single subsystem decision and generate a conclusion. This is the true reason that multimodal biometric systems are more accurate than unimodal or any other traditional authentication system.
b) Increased and Reliable Recognition:A multimodal biometric system authorized a greater level of confidence for an accurate match in identification as well as verification modes. As multimodal biometric systems uses multiple biometric traits, each single biometric trait can offer additional proof about the authenticity of any user’s claim. For example, the patterns of movements (gait) of two persons of the same family or by coincidentally of two different persons can be similar. In this particular situation, a unimodal biometric system that depends only on gait pattern analysis might lead to a false recognition. If the same biometric trait additionally includes fingerprint matching then the system would absolutely results in increased recognition rate, because it is nearly impossible that two different persons have same gait as well as fingerprint pattern.
c) Enhanced Security: In Multimodal biometric system, use of multiple methods of identification and verification, a system can acquire higher degree of threshold recognition and a system administrator can get a decision on the accurate level of security that is important. For an extremely high security point of view, you can use up to three biometric identifiers of an individual and for a lower security point of view; you could possibly require one or two biometric identifiers. If one of the attribute fails for any unknown reason although your system can still use another one or two of them in order to provide the accurate identification and verification of a person.
d) Vulnerability: Spoof attacks are the biggest threat to authentication systems. Unimodal as well as Multimodal biometric systems are sometimes vulnerable to spoof attacks. Spoofing occurs whenever an unauthorized user has the capacity to authenticate as an authorized user. The potential threats due to artificial or fake fingers were examined by another experiment and researcher’s team as exhibited that artificial fingers made with plastic molds could possibly enroll in the 11-12 tested fingerprint systems and were being accepted in the identification and verification procedures with the major probability of 68-100%, depending on the biometric system. In this case, alternative hardware device that depends on simultaneous multimodal authentication such as a biometric smart fingerprint scanner with liveness detection with the help of temperature sensing can remove spoofing. “Liveness” can be describes the capacity of a multimodal biometric system to distinguish between a fake and a living sample and is generally done by measuring biometric features like humidity, blood flow, pulse, temperature, etc.
e) User Acceptance:As already mention above multimodal biometric systems are more reliable, accurate, ability to avoid spoofing attacks, have good security options, and this type of systems are more widely accepted by many countries. However, in deployments of large countries where accuracy and security are paramount, no matter how small size, multimodal systems have become ubiquitous and necessary .
2. BIOMETRIC SYSTEM MODULES
Generally biometric system can be viewed as having four modules: a sensor module; a quality assessment and feature extraction module; a matching module; and database module as shown below figure:
1:1 & 1:N matching
Genuine or imposter and identity / reject
a) Sensor module: A suitable biometric sensor is required to acquire the raw biometric data of an individual. For example, to acquire fingerprint images, an optical fingerprint sensor may be used to acquire ridges structure of the fingertip.
b) Quality assessment and feature extraction module: The quality of the biometric data acquired by the sensor is assessed in order to determine its suitability for further processing. Sometimes, the quality of the acquire data may be so poor that the user is asked to present the biometric data again. The biometric data is then processed and a set of salient features extracted to represent the identity. For example, the position and orientation of minutia points (ridge and valley anomalies) in a fingerprint image are extracted by the feature extraction module in a fingerprint-based biometric system. During enrolment process, this feature set is stored in the database and is commonly known as a template.
c) Matching module: The extracted features are now compared against the stored templates in database to generate match scores. In a fingerprint-based biometric system, the no. of matching minutiae between the input and the stored template feature sets is computed and a match score generated. It may be one to one matching that is verification or one too many matching that is identification.
d) Decision module: Decision module generates output based on match score as in matching module. In verification process, the output is either imposter or genuine. In identification process, the output is either identity or reject. In both cases the low match score is emitted to the system .
2.1 Fusions in multimodal biometric
Fusion is a mechanism that can combine the all classifications of result from different media/channels. Multimodal biometric system can acquire the two or more biometric traits to increase its strength due to present multiple traits and difficult to forge. The performance of the multimodal system is examined in term of image acquisition errors and matching errors. Matching errors are false match rate (FMR), in which an impostor user’s sample matches with a legitimate user’s template, and False Non Match Rate (FNMR), in which a legitimate user’s samples don’t match her/his own template. Image acquisition errors are Failure to Enrol (FTE) which is defined as a person unable to successfully enrol himself in a biometric system and Failure to Acquire (FTA) in which a person unable to provide a good quality of biometric trait at verification and identification .
There are various levels of fusion as defined below:
2.1.1 Sensor level fusion
At sensor, the raw data acquired from multiple sensors (for e.g. one sample from optical sensor and other sample from solid state sensor) and from multiple traits (for e.g. one sample of face and another sample of fingerprint) and combine to generate a new biometric data from which a feature set can be extracted.
Sample 1 Accept
Sample 2 Reject
Fig 3: Sensor Level Fusion
Figure 3 indicate, sensor level fusion as different biometric samples as inputs in sensor. Where FM: Fusion Module, FE: Feature Extraction module, MM: Matching Module, DM: Decision Module, DB: Database.
2.1.2 Feature extraction level fusion
At feature level, a set of features are extracted from different biometric traits can be fused by using a specific fusion algorithm and to form a composite feature set of that biometric traits. For e.g. a feature set of fingerprint and face can be combined. Fusion at Feature level is better than other fusion level and to increase accuracy of result.
Image 1 Accept
Image 2 Features Reject
Fig 4: Feature level Fusion
Figure 4 indicate the feature extraction level fusion where both channels feature set are fused for further processing. Where FM: Fusion Module, FE: Feature Extraction module, MM: Matching Module, DM: Decision Module, DB: Database.
2.1.3 Match score level Fusion
At matching module, different sample’s feature set is compared with the templates stored in the database and a match score is generated. After match score generated, fusion is apply and both match scores are combined to generate a new match score and then this new match score send to decision module.
Score values Accept/Reject
Fig 5: Match score level fusion
Figure 5 indicate match score level fusion and send this score to decision module. Where FM: Fusion Module, FE: Feature Extraction module, MM: Matching Module, DM: Decision Module, DB: Database.