Source-linked AI summary
An Efficient Secure Multimodal Biometric Fusion Using Palmprint and Face Image
M. Nageshkumar, P. K. Mahesh, M. N. S. Swamy
TL;DR
Because a single biometric feature can be insufficient for identity verification, the paper combines face and palmprint recognition using matching-score fusion. It reports accuracy above 98% and better performance than unimodal biometrics.
Abstract
from arXiv · showhide
Biometrics based personal identification is regarded as an effective method for automatically recognizing, with a high confidence a person's identity. A multimodal biometric systems consolidate the evidence presented by multiple biometric sources and typically better recognition performance compare to system based on a single biometric modality. This paper proposes an authentication method for a multimodal biometric system identification using two traits i.e. face and palmprint. The proposed system is designed for application where the training data contains a face and palmprint. Integrating the palmprint and face features increases robustness of the person authentication. The final decision is made by fusion at matching score level architecture in which features vectors are created independently for query measures and are then compared to the enrolment template, which are stored during database preparation. Multimodal biometric system is developed through fusion of face and palmprint recognition.
1 INTRODUCTION
A single biometric feature can be insufficient for reliable identity verification, motivating multimodal systems that combine traits. The paper proposes face-and-palmprint fusion and notes that earlier studies used holistic face features on small datasets.
1 INTRODUCTION
- Combining biometric modalities is proposed to improve verification when a single feature is not exact enough.The introduction frames face and palmprint acquisition as non-invasive and low-cost.Source: §1 · p. 1
- Earlier face-and-palmprint studies used holistic face features and reported results on small datasets.Source: §1 · p. 1
- The introduction identifies fusion as necessary to integrate information from the separate biometric modalities.Source: §1 · p. 1
- The paper presents a face-and-palmprint fusion strategy at the feature level.Source: §1 · p. 1
2 SYSTEM STRUCTURE
The system processes face and palmprint traits separately, combines their matching scores, and bases the final decision on a threshold.
2 SYSTEM STRUCTURE
- Face and palmprint features are extracted and matched independently before their scores are summed for the final threshold decision.The system structure is shown in Figure 1.Source: §2 · p. 2, Figure 1 · p. 2
3 FEATURE EXTRACTION USING CANONICAL FORM BASED ON PCA APPROACH
The feature-extraction approach uses PCA-related eigenvalues and eigenvectors to represent face and palmprint images, then applies a canonical form. The section also describes distance-based matching and weighted score fusion.
3 FEATURE EXTRACTION USING CANONICAL FORM BASED ON PCA APPROACH
- The training-image average is defined before PCA is applied to image deviations.The images are represented as vectors, and each image differs from the average by a vector.Source: §3 · p. 2
- The canonical feature method represents image differences using principal components and eigenvalues of the image covariance matrix.The paper describes eigenvectors as features that characterize differences between images.Source: §3 · p. 2
- The canonical form is also called the sum-of-squares form or principal-axes form.The described extraction steps include determining image eigenvalues and eigenvectors, then using the canonical form.Source: §3 · p. 3
- The procedure preprocesses only palm images before determining eigenvalues and eigenvectors and extracting canonical features.Source: §3 · p. 3
- The final matching score is a weighted sum of face and palm scores, compared with a threshold to classify a person as genuine or impostor.The section also introduces Euclidean distance between feature-vector instances as a matching measure.Source: §3 · p. 3
5 EXPERIMENTAL RESULTS
The evaluation uses face-and-palmprint image pairs from 120 subjects.
5 EXPERIMENTAL RESULTS
- The evaluation dataset contains 720 pairs of images from 120 subjects, with face and palmprint images in the training database for each subject.Source: §5 · p. 3
4 FUSION
The fusion experiment combines face and palmprint matching scores using a sum-of-scores technique. The reported overall accuracy exceeds 97%, with FAR of 2.4% and FRR of 0.8%.
4 FUSION
- The fusion procedure extracts features from the query image, calculates weights α and β, then sums the face and palmprint matching scores.The resulting final score is passed through the fusion process described for the two traits.Source: §4 · p. 3
- The palm and face image figures each show five stages: original, grayscale, resized, normalized modal, and diagonalization images.The captions identify the corresponding palm and face processing sequences.Source: Figure 2 · p. 4, Figure 3 · p. 4
- The individual biometric systems were first developed and tested using FAR, FRR, and accuracy.Table 1 is identified as reporting those measures for the systems.Source: §4 · p. 4, Table 1
- Overall accuracy exceeded 97%, with FAR of 2.4% and FRR of 0.8% for the combined system.The reported result follows fusion of both traits at matching-score level using the sum-of-scores technique.Source: §4 · p. 4
6 CONCLUSION
The conclusion combines face and palmprint scores at classifier and trait levels to address cases where one biometric trait's data is unavailable. It reports multimodal accuracy above 98%, better than unimodal biometrics.
6 CONCLUSION
- The conclusion reports multimodal accuracy above 98%, outperforming unimodal biometrics.The paper describes combining individual face and palmprint scores at classifier and trait levels.Source: §6 · p. 4
- The system combines individual face and palmprint scores at classifier and trait levels to form a multimodal biometric system.Source: §6 · p. 4
- The conclusion identifies missing biometric data for a particular trait as a failure case for unimodal systems.Source: §6 · p. 4