^{1}

^{*}

^{2}

This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

Biometric systems based on uni-modal traits are characterized by noisy sensor data, restricted degrees of freedom, non-universality and are susceptible to spoof attacks. Multi-modal biometric systems seek to alleviate some of these drawbacks by providing multiple evidences of the same identity. In this paper, a user-score-based weighting technique for integrating the iris and signature traits is presented. This user-specific weighting technique has proved to be an efficient and effective fusion scheme which increases the authentication accuracy rate of multi-modal biometric systems. The weights are used to indicate the importance of matching scores output by each biometrics trait. The experimental results show that our biometric system based on the integration of iris and signature traits achieve a false rejection rate (FRR) of 0.08% and a false acceptance rate (FAR) of 0.01%.

Multi-modal biometric systems address the shortcomings of uni-modal systems. For instance, the problem of non-universality: it is possible for a subset of users to not possess a particular biometrics trait. For example, the feature extraction module of an iris authentication system may be unable to extract features from iris images associated with specific individuals, due to either the occlusion of the iris region of interest or poor quality of the images. Multi-modal systems ascertain that a

Furthermore, multi-modal biometric systems are expected to be more reliable due to the presence of multiple pieces of evidence [

In this paper, a framework for modeling bi-modal biometric systems based on iris (a physiological trait) and the signature (a behavioral trait) for personal authentication is proposed. These two biometric traits are not correlated. Moreover, iris is proving to be one of the most reliable biometric traits while signatures continue to be widely used for personal authentication.

Multi-modal biometrics was pioneered by Anil K. Jain; and there has been substantial research carried out in this area. A variety of biometric fusion schemes, which use classifiers, have been described in the literature to combine multiple biometric trait scores. These include majority voting, sum and product rules, k-NN classifiers, SVMs, and decision trees [

Other multi-modal biometric fusion approaches include: the HyperBF network approach used to combine the normalized scores of five different classifiers operating on the voice and face feature sets of an individual for identification [

Although several score fusion techniques have been proposed in the literature, Ross

_{1}, _{2}, …, _{n}

_{1}, _{2}, …, _{n}

In this paper, an enhanced user-specific weighting technique is proposed, which is based on the different degrees of importance for different traits of an individual to integrate the physiological trait, the

The rest of the paper is structured as follows: Section 3 explores various fusion techniques for combining biometric traits; Section 4 describes an overall multi-modal biometrics system; Section 5 describes the weighting techniques and normalization strategies; Section 6 presents experimental results; and Section 7 draws the conclusions and future work.

Multi-modal biometric systems are based on the consolidation of information presented by multiple evidences that stem from multiple traits. Some of the limitations imposed by uni-modal biometric systems (that is, biometric systems that rely on the evidence of a single biometric trait) can be overcome by using multiple biometric modalities [

A variety of factors should be considered when designing a multi-biometric system. These include the choice and number of biometric traits; the level in the biometric system at which information provided by multiple traits should be integrated; the methodology adopted to integrate the information; and the cost

A simple multi-modal biometrics system has five important components as depicted in

There are various levels of fusion for combining biometric traits. The three possible levels of fusion are [

A brief description of the two biometric traits used in this research work is given below.

Iris recognition is proving to be one of the most reliable biometric traits for personal identification since iris patterns have stable, invariant and distinctive features. Several techniques have been proposed for iris segmentation, coding and matching. The most common approach used in iris recognition is to generate feature vectors corresponding to individual iris images and perform iris matching based on some distance measures [

Signature continues to be an important biometric trait because it remains widely used primarily for authenticating the identity of human beings. An efficient text-based directional signature recognition algorithm which verifies signatures, even when they are composed of symbols and special unconstrained cursive characters which are superimposed and embellished is used [

The iris and signature traits are fused at the matching score level, where the matching scores output of each of these two traits are weighted and combined. Fusion at the matching score level is usually preferred, as it is relatively easy to access and combine the scores presented by the different modalities [

_{iris}

_{sig}

_{iris}

Iris matching scores are computed from string iris feature codes extracted by the cumulative-sum-based grey change analysis technique. To verify the similarity of two iris codes, Hamming Distance (HD) based on the matching algorithm [_{iris}

where _{h}_{v}_{h}_{v}

Signature matching scores are generated from the signature feature vectors. To verify the similarity of two signatures, Mahalanobis Distance (MD) based on correlations between signatures is used. It differs from Euclidean distance in that it takes into account the correlations of the data set and is scale-invariant. The smaller the MD, the higher the similarity of the compared signatures. The MD denotes the signature raw matching score, _{sig}

where

Given a set of _{k}

_{k}_{k}_{k}

_{k}

_{k}

The ROC curves depicting the performance of the individual score normalization techniques implemented on iris biometrics trait is shown in

Let
_{fus}

where _{iris}_{sig}

Different iris scores and signature scores are given different degrees of importance for different users. For instance, by reducing the weight _{iris}_{sig}

Let
^{th}

For the ^{th}^{th}

Choose that set of weights that minimizes the total error rate. The total error rate is the sum of the false acceptance and false rejection rates pertaining to this user.

The set of weights,
^{th}

Let
^{th}_{1} and _{2}

and

where
_{1} and _{2}. Then, the fusion weights for the ^{th}

The dual _{iris}_{sig}_{iris}_{sig}_{iris}

and the weighted signature score _{sig}

The weighted matching scores and their labels are used to train the 2

and

where

where _{+}_{−} are the number of genuine and impostor, respectively. The training data is mapped into a higher dimension feature space such that

In the classification phase, the bi-modal fusion matching score _{fus}

where

where _{iris}, a_{sig}, b_{iris}_{sig}_{fus}

The performance of the investigated bi-modal biometrics system is evaluated by calculating its false acceptance rate (FAR) and false rejection rate (FRR) at various thresholds. These two factors are integrated together in a receiver operating characteristic (ROC) curve that plots the FRR or the genuine acceptance rate (GAR) against the FAR at different thresholds. The FAR and FRR are computed by generating all possible genuine and impostor matching scores and then setting a threshold for deciding whether to accept or reject a match.

The bi-modal database used in the experiments was constructed by merging

Firstly, the matching scores of the iris and signature traits are computed as defined in

The ROC curves in

The user-score-based weighting algorithm computes the weights of the iris and signature traits by analyzing how close the two matching scores are to their respective thresholds, hence associating the weights with the different degrees of importance for the bi-modal biometric traits involved. Comparatively, the exhaustive search weighting technique calculates weights that simply minimize the total error rate. This minimum error rate (the sum of FAR and FRR) does not necessarily reflect the different degrees of importance for the bi-modal biometric traits fused.

In this paper, an enhanced user-specific weighting technique of integrating a physiological biometrics trait, the

Portions of the research in this paper use the CASIA iris image database collected by the Institute of Automation of the Chinese Academy of Sciences, and the

Multi-modal Biometrics System (Iris & Signature).

ROC curves showing the performance of each of the three normalization techniques on the Iris trait.

Average true positive rate of the iris and signature Modalities.

Tanh normalized-based ROC curves showing the performance of using Iris, Signature, Iris + Signature (Exhaustive), and Iris + Signature (User-score-based).

User-specific Scores and Weights of different traits for 10 users.

1 | 0.192 | 0.001 | 0.487 | 0.488 | 0.80 | 0.20 |

2 | 0.277 | 0.001 | 0.490 | 0.488 | 0.86 | 0.14 |

3 | 0.625 | 2.054 | 0.505 | 0.505 | 0.50 | 0.50 |

4 | 0.446 | 2.438 | 0.506 | 0.496 | 0.44 | 0.56 |

5 | 0.232 | 0.005 | 0.486 | 0.492 | 0.83 | 0.17 |

6 | 0.473 | 2.383 | 0.498 | 0.507 | 0.47 | 0.53 |

7 | 0.071 | 0.028 | 0.484 | 0.493 | 0.67 | 0.33 |

8 | 0.522 | 2.474 | 0.505 | 0.507 | 0.47 | 0.53 |

9 | 0.366 | 1.358 | 0.497 | 0.502 | 0.48 | 0.52 |

10 | 0.451 | 1.774 | 0.502 | 0.506 | 0.50 | 0.50 |

Exhaustive search

Exhaustive search | 0.01 | 0.75 |

User-score-based | 0.01 | 0.08 |

Comparative table of the weighted based fusion algorithms.

Face + Iris | Quality based Sum-rule [ |
97.39 |

Face + Speech | k-NN based fusion [ |
99.72 |

Face + Iris | Quality based [ |
98.91 |

Iris + Signature | User-Score-based Weighted 2 |
99.6 |