The History Of Multimodal Biometric System

Published Date: 02 Nov 2017

Establishing person identity is a critical issue to a wide variety of applications such as access control, electronic commerce, banking, communications etc. Knowledge based methods like pass words or token based methods like ID cards, Pan cards etc., can be stolen, shared or forgotten. Biometric person identification is described as identifying a person based on his/her physical or behavioral characteristics. A biometric system verifies the user characteristics like face, finger print, palm, iris, voice, gait, signature etc., for a reliable authentication system. Unimodalsystem uses a single modality and multimodal system uses multiple characters of a person in establishing the identity. Input identity is verified against stored template in the data base. Biometric system performs in four levels of operation. Sensor level which captures the input data, feature level which extracts the salient features of the captured image, score level which finds the match score by comparing it with the stored template in the data base and decision level which helps in establishing the identity of the input image. The thesis is proposed to implement unimodal and multimodal biometric system.

http://t2.gstatic.com/images?q=tbn:ANd9GcQxKDg97EqiH2MmK22xL-K6qX1rWErmME2hy2EHyF-q_Qb5CWhR

Fig1 Biometric system

Biometrics is measurements of the characters of a personthat are used for identification. Measurements related to each character will differ. Biometric system is analyzed based on cost, accuracy, complexity, inter-operability and system security. Improvement in performance of a biometric system can be achieved using various approaches of feature extraction, using multiple classifiers, multiple samples of the trait etc. Various standard like NISIT/ANSI, ISO/IEC, were set for the measurements and analysis of a biometric system. Performance of a biometric system is analyzed with error estimation like false match, false mismatch, imposter acceptance rate, ROC, DET, plots. Analysis of a biometric system and its performance is presented in the thesis work by addressing various issues such as biometric limitations, biometric scenario, biometric system requirement and biometric security in chapter 2 of the thesis report. With help of the analysis, palm print and face is chosen as a biometric character for their rich feature information and are used in implementation of unimodal and multi modal biometric recognition systemin the thesis work.

Palm print recognition has emerged as highly accepted biometric system due to its easy acquisition and highly reliable. The inner surface of palm contains principal lines, wrinkles, singular points, ridges and minutiae. A region of interest (ROI) is extracted from the palm area for processing. Palm recognition process includes feature extraction (stored as template in the data base) matching (input query features are matched with stored features) and decision making(to accept or reject the query based on match score).The thesis work is carried out by analyzing the palm print recognition system, palm features, its processing levels, data bases available etc. The analysis is presented in chapter 3 of thesis report. The implementation of a palm print biometric recognition system is carried out using edge detection techniques. Palm print recognition system processing stages is shown in the block diagram. Sobel, canny and multiscale edge detection techniques are used for feature extraction. The feature extraction results are presented in the chapter4 of the thesis report.

C:\Documents and Settings\seshikala\Desktop\kiran pictures modified\4.png

Face is chosen as other biometric character for implementation of a biometric system. Face is most common biometric used by humans to recognize people. Face recognition has received substantial attention from researchers due to its applications of security like airport, criminal detection, face tracking, forensic, surveillance systems etc. Compared to other biometric traits like palm print, Iris, finger print etc., face biometrics can be non-intrusive and can be taken even without users knowledge

C:\Documents and Settings\seshikala\Desktop\kiran pictures modified\5.png

Face biometrics is a challenging field of research with various limitations imposed for a machine recognition like variations in head pose, change in illumination, facial expression, aging, occlusion due to accessories etc,. Various approaches were suggested by researchers in overcoming the limitations stated. In the thesis work Face Recognition is implemented using curvelet transform for feature extraction and SVM for classification.

Image representation should satisfy the condition of multi resolution, localization, critical sampling, directionality and anisotropy. Curvelets which is a multi-scale, multi-resolution transform overcomes the limitations of dimensionality problem and provides optimal representation of objects with curve singularities. Curvelet uses only small number of coefficientsto represents a line or a curve in a given image. Face recognition system using curvelet transform is implemented in the thesis work.

Unimodal biometric system is subjected to various limitations like noisy sensor data, spoof attack, inter class similarity which can be overcome by using Multimodal biometric system.Multimodal biometrics integrates information from various modalities to overcome various limitation of unimodal biometric system. Multimodal biometric system can be implemented using any of the four levels of fusion in sensor level, feature level, match core level and decision level. The fusion scenarios and strategies were studied and presented in the thesis for selection of fusion level for palm and face. The thesis work is carried by selecting feature level fusion for integrating the information of palm and face. The analysis of feature level is given in the chapter 6 of the thesis report.

http://www.zemris.fer.hr/projects/Biometrics/slike/sl1-s.jpg

Fig 3 . Multimodal Biometric System

Features sets of each biometric are generated separately and are fused together to produce a single multimodal biometric template. The matcher uses this template to generate a match score and a final decision making. Feature sets consists of rich source of information of the input data and fusion at feature level is expected to give more efficient recognition compared to other levels of fusion. Feature level fusion is an understudied problem by researchers when compared to the other levels of fusion due to the limitations stated below.

Feature vectors of different biometric traits may exist in different domain and thus becomes incompatible for fusion.

Relationship between the feature spaces of different biometric traits used may not be known

If features are compatible and fused the resulting feature set may have a large dimension and requires feature reduction techniques to reduce the feature set

The problem of dimensionality in feature level fusion is always addressed by researchers as "curse of dimensionality". Feature reduction may result into loss of information in feature set and reduce the accuracy of the system. The face and palm features are compatible in feature space and are usedfor feature level fusion. The Palm print recognition system using curvelet transform is implemented. The result of face recognition is used for fusion.The block diagram of the implementation is show in fig 4.Fusion is carried out before matching. Curvelet transform feature extraction technique is used for both unimodal and multimodal recognition system implementation. The result of the implementation and its comparison with unimodal is given in chapter7 of the thesis.