User:Cadhawale/sandbox

Multimodal Biometrics

In an identity management system like banking site log in, ATM login, Airport verification for a traveler, traditional token based user verification methods based on license, passport, PINS or passwords prove as weak authentication methods since they can be missed or stolen. Authentication measure is thus required to exist with user, not lost in any case and must be unique of course! Such a reliable measure is nothing but the person itself!, Here the term biometrics came into existence.

The term Biometrics contains two words − Bio (Greek word for Life) and Metrics (Measurements). Biometrics is a branch of information technology that aims towards finding one’s identity based on personal traits. Biometric systems are developed based on the individual’s physical characteristics or behavioral characteristics. The physical characteristics or features or attributes like finger prints, color of iris, color of hair, hand geometry, and behavioral characteristics such as tone and accent of speech, signature, etc. make a person stand separate from the rest. These are the unique characteristics which are selected. This means that we should have feel of uniqueness when we look at the object. Not only one part should be different but together the entire object must be found unique. Also it is necessary that it should be existing  with everybody. Biometric systems use these unique features for − •	Identification and verification of a person •	Authentication of person •	Security of  system   from un-authorised  access. A biometric system is a technology which takes input as person’s  physiological, behavioral, or both traits  as input, analyzes it, and authenticate the person for accessing the system.

NEED OF MULTIMODAL BIOMETRICS

Single feature selection for biometric recognition can lead to wrong identification of the objects in rare cases where dirty data acquisition is done or the cases where people are trying to crack the system by proxy input. It has also the drawback of non-availability of universal biometric traits. Also in this case the issues of interclass-variations (variation in biometric traits due to age) and intra-class similarities (twins with similar physical characteristics) hold true. In this case multimodal biometric system is used, for example use of (Photograph+signature) for person identification. Multimodal biometrics does not always refer to the use of two or more separate biometric trait samples but the multiple inputs can come from variety of sources (Andreas Humm, 2009). These sources are summarized in Figure 1.The need of multiple attributes vary with applications. For example, employee attendance system require person’s unique thumb print, while detecting a criminal running among the crowd at a public place requires walking style and body silhouettes to be traced and matched. In a vehicle license management system face and signature of the driver may work whereas in online course tutorial conduction student’s keystroke dynamics can play a good part to find him unique.

REVIEW OF MULTIMODAL BIOMETRICS

In multibiometric systems the biometric traits are fused at different level like feature level, match score level, decision level, etc. with the help of various fusion methodologies; some of which can be listed as concatenation, weighted summation, product, min, max, borda count, majority voting etc. following section summarizes work by different authors on Multibiometric systems with respect to modalities, fusion ways, fusion levels, database, dataset size, algorithms, performance, etc.

Multibiometric Fusion Ways Different modalities in Multibiometrics are fused in different ways at various levels right from input acquisition to decision making. Those are summarized below- i>	Single Biometric Multiple Sensors In these systems, a single biometric trait is imaged using multiple sensors in order to extract diverse information from the images. For example, the system may record the 2-D texture content of a person’s face using a CCD camera and the 3-D surface shape of the face using a range sensor in order to perform authentication. The availability of multi-sensor data pertaining to a single trait can assist the segmentation and registration procedures also besides improving matching accuracy (Ross, 2006). In the literature pioneer work on multiple-sensors is given about multisensory fingerprint verification (Marcialis & Roli, 2004) where the system is based on the fusion of optical and capacitive sensors.

ii>	Multiple Biometrics Under these systems combination of different physical\behavioral biometrics can be used to identify\verify accurate users. For ex. Some of the earliest work has used face and speech (Souheil Ben-Yacoub, 1999). Physically uncorrelated traits handwriting and speech (Andreas Humm S. M., 2009) are found to give superior results than that of correlated traits palm print and fingerprint (Yong Jian Chin, 2009).

iii>	Single Biometric Multiple Units/Instance The input samples are multiple instances of same trait. For ex. use of left and right index fingers. In this case an individual’s own biometric data can be used to verify its own identity. For ex. If a person’s single unit is found badly captured due to some inherent problems like skin decease, then the combination of facts across multiple will become a good identifying feature. For big number of subjects in database, multiple instance systems are more useful (Ross, 2006). For example Ajay Kumar (Ajay Kumar, 2011) has proposed fusion of multiple representations of Palmprint. iv>	Multiple Algorithms/Classifier/Matcher These systems are cost-effective and user-convenient, since they do not incorporate usage of different sensors; instead makes use of various matcher algorithms for different features. For example for the same fingerprint image different algorithms for minutiae extraction and ridges separation. For example in the work on verification for the feature of finger knucle (AlMahafzah, Imran & Sheshadri, 2012) the authors have used multiple algorithms like Log-Gabor Filters (LG), Local Phase Quantization (LPQ), Principal Component Analysis (PCA) and Locality Preserving Projections (LPP).

v>	Hybrid Systems These are the biometric systems where combined approaches using any of the types stated above. For ex. Ramli et al. (Ramli, Rani, & Ishak, 2011) provide solution to identify a person by making use of multiple instances of speaker output combined with face modality by employing weighted sum rule fusion and Min-max technique for normalization. Thus the system is multi-instance and multimodal in its design.