Authors:
Mahrukh Siddiqui、Shahzaib Iqbal、Bandar AlShammari、Bandar Alhaqbani、Tariq M. Khan、Imran Razzak
Paper:
https://arxiv.org/abs/2408.09426
Introduction
In recent years, contactless biometric technologies have garnered significant attention due to advancements in sensing technologies and their applications in commercial sectors. The National Institute of Standards and Technology (NIST) has emphasized the development of Next Generation Fingerprint Technologies, highlighting the potential of contactless fingerprint systems. Unlike traditional contact-based fingerprint systems, contactless systems offer several advantages, including reduced contamination, minimized non-linear distortion, and improved hygiene. However, contactless fingerprint images present unique challenges, such as low ridge/valley contrast and the absence of elastic deformation, which complicate the enhancement and matching processes.
This study introduces a novel algorithm designed to enhance and match contactless fingerprint images. The proposed method aims to improve the accuracy of minutiae detection through enhanced frequency estimation and a region-quality-based minutia extraction algorithm. Additionally, a highly accurate minutiae-based encoding and matching algorithm is presented. The effectiveness of the proposed approach is validated through extensive experimental testing, demonstrating superior performance compared to existing state-of-the-art techniques.
Related Work
Spatial Domain Filtering
Spatial domain filtering techniques have been widely used to enhance fingerprint images. Contextual filters, introduced by Nickerson and O’Gorman, are commonly employed to improve fingerprint images by regulating ridge frequency and orientation. However, these methods often assume a constant local ridge frequency, leading to imperfect filtering in some areas. Hong et al. proposed an efficient enhancement technique based on the Gabor filter, which adapts to local ridge orientations and frequencies. Despite its advantages, this method performs sub-optimally in areas where ridge patterns deviate from a pure sinusoidal pattern.
Fourier Domain Filtering
Fourier domain filtering techniques, such as those proposed by Sherlock et al., improve fingerprint images by applying the fast Fourier transform. These methods involve multiplying the Fourier transform of the fingerprint image by precomputed filters. However, the constant ridge frequency assumption remains a limitation. Yin et al. introduced a new technique focusing on enhancing contactless fingerprints, addressing some of the unique challenges posed by contactless fingerprint images.
Research Methodology
Fingerprint Enhancement
The proposed algorithm consists of two main phases: offline and online. The offline phase involves enhancing fingerprint images, extracting minutiae features, and encoding them to create an encoded candidate database. The online phase includes acquiring a new template image, enhancing it, extracting minutiae features, and encoding it for matching against the pre-encoded candidate database.
Region of Interest (ROI) Segmentation
The first step in fingerprint enhancement is segmenting the fingerprint image from the background to obtain the region of interest (ROI). This is achieved using the average magnitude of the gradient, which is higher in the foreground and lower in the background.
Gabor-Based Contextual Filtering
Local ridge orientations and frequencies are calculated to apply Gabor-based contextual filtering. The orientations are computed using the inverse tangent of the gradient in both horizontal and vertical directions. The local ridge frequencies are estimated using a modified version of the x-signature method, which provides accurate estimations even in the presence of high ridge curvature.
Binarization and Thinning
The enhanced image undergoes binarization and thinning to produce a ridge/valley skeleton. A quality mask is created to extract high-quality minutiae features from the thinned image, leading to accurate fingerprint identification and reduced computational complexity during feature matching.
Feature Extraction
Minutiae points, such as ridge endings and bifurcations, are extracted from the thinned fingerprint image. A local neighborhood of each pixel is scanned within a 3 × 3 window to classify ridge pixels based on the number of transitions from 0 to 1. False minutiae are removed to obtain a reliable list of minutiae points.
Feature Encoding
Each minutiae point is selected as a reference point and encoded based on its neighboring minutiae. The encoding process involves calculating the relative distance and angle features between the focal minutiae and its neighboring minutiae. The final list of encoded minutiae is stored in the database as a candidate finger-code.
Feature Matching
To match a candidate fingerprint image with template images stored in the database, an exhaustive search algorithm is employed. The biometric code of each minutiae point in the candidate image is matched with all minutiae codes in the template image to find the best match. A similarity score is calculated based on the matching pairs of minutiae between the two fingerprint images.
Experimental Design
Dataset and Evaluation Protocol
The PolyU contactless fingerprint database is used for experiments, following the FVC testing protocol. The dataset consists of 336 subjects, each with 6 samples. The equal error rate (EER) is used as the performance indicator, calculated in two stages: false non-matching rate (FNMR) and false matching rate (FMR).
Experimental Stages
- Genuine Matching: Each sample of a subject is matched with the remaining samples of the same subject to compute the FNMR.
- Imposter Matching: The first sample of each subject is matched with the first sample of the remaining subjects to compute the FMR.
Performance Metrics
The EER is a measure of the system’s performance, representing the point where both FMR and FNMR are equal. The experiments involve a total of 10,080 genuine matching and 56,280 imposter matching experiments.
Results and Analysis
Enhancement and Minutiae Extraction
The proposed method demonstrates significant improvements in the enhancement and minutiae extraction stages. The enhanced images undergo binarization and thinning, followed by minutiae extraction, as illustrated in Figure 4.
Matching Performance
The proposed method achieves a minimum EER of 2.84% for 9 closest neighbors and a matching neighbor threshold of 5. Table I presents the empirical analysis of EER, showing that increasing the number of neighbors involved in encoding improves the accuracy of the matching algorithm. Table II compares the EER of the proposed method with other matching algorithms, demonstrating superior performance.
Overall Conclusion
This study presents a robust algorithm for contactless fingerprint enhancement and matching, addressing the unique challenges posed by contactless fingerprint images. The proposed method improves the clarity of pertinent details while mitigating noise artifacts, leading to accurate identification. The experimental results validate the significant superiority of the proposed approach, achieving a minimum EER of 2.84% on the PolyU contactless fingerprint database. Future research directions may include GAN-based fingerprint enhancement and the integration of GAN with existing forensic tools for seamless adoption in real-world applications.
Figure 1: Example images of contactless and contact-based fingerprints.
Figure 2: Pipeline of the proposed fingerprint recognition system.
Figure 3: Fingerprint matching based on neighboring Minutia.
Figure 4: Visual performance of the proposed method. From left to right: input image, the corresponding mask, binary image, thinned image, and minutiae marked on thinned image.
Table I: EER comparisons of different n-nearest neighbors for different matched neighbors on the PolyU contactless fingerprint database.
Table II: EER and time comparison of the proposed method with other matching algorithms on the PolyU contactless fingerprint database.