A hybrid proposed image quality assessment and enhancement framework for finger vein recognition

• Published in: Multimedia tools and applications 2024-02, Vol.83 (5), p.15363-15388
• Main Authors: Shaheed, Kashif, Qureshi, Imran
• Format: Article
• Language: English
• Subjects: 1158T: Role of Computer Vision in Smart Cities: Applications and Research Challenges; Algorithms; Biometric recognition systems; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Fingers; Image classification; Image contrast; Image enhancement; Image quality; Multimedia Information Systems; Quality assessment; Special Purpose and Application-Based Systems; Veins
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-021-11877-x

Finger vein recognition (FVR) is a biometric trait that can authenticate the person in real-time applications. However, finger vein (FV) images are generally poor due to various unfavorable factors. Therefore, these images are prone to low contrast, insufficient brightness, and noise problems, significantly impacting the FVR system performance. Hence, image quality assessment and enhancement play a vital role in finger vein recognition. To advance the FVR system performance, we propose two FVR algorithms based on FV image quality estimation and an enhancement algorithm. At first, good quality feature such as contrast, entropy and information capacity were extracted from finger vein images. Then, the image quality is evaluated by the KNN with the r-smote technique to classify the FV image into two classes, High Quality (HQ) and Low Quality (LQ) images. Second, a novel enhancement method called guided filter and bilateral filter (GFBF) are presented to enhance the low-quality FV images. Afterward, we estimate the enhancement algorithm by using SSIM and PSNR. Finally, we evaluate and test the proposed system strength using two parameters—namely accuracy and equal error rate (EER), for Classifier and recognition performance, respectively, on a dataset of 1052 FV images. The completed experiment determined that the proposed image assessment and enhancement method outperformed other enhancement and assessment schemes by achieving a low identification error rate of 0.0335. Further results conclude that the proposed art would be a perfect pre-processing tool for finger vein feature-based algorithms.