Scholarlink

<p>Fingerprint recognition is essential element of biometric security, yet it faces challenges related to accuracy, robustness, and latency. One significant issue is that the performance of fingerprint authentication systems significantly decreases when faced with low-quality fingerprints. In this paper, two new hybrid approaches are suggested to enhance the accuracy of fingerprint verification and create a reliable identification system. The first approach, SIFT-Descriptor-SCNN, combines the scale-invariant feature transform (SIFT) method with the proposed Siamese convolutional neural network model. First, we compute the SIFT feature descriptors and then pass them to the SCNN for further feature extraction and matching. This increases the system’s ability to distinguish between various patterns for accurate identification. The second novel approach combines oriented FAST with a dense grid, the SIFT algorithm, and the SCNN, named oriented FAST-Dense-SIFT-Descriptor-SCNN. In this work oriented FAST with a dense method is implemented for the detection of keypoints and then the detected keypoints are fed into SIFT algorithm to compute the descriptors. The integration of these algorithms with the SCNN model enhances the performance of the fingerprint identification system. We test the performance of the two hybridization models, the SOCOFing dataset, which includes 6000 real fingerprint and 49,270 synthetically altered versions organized into easy, medium and hard was used. The first model provides an accuracy of 97.57, 96.72, and 95.61 for the easy, medium, and hard datasets, while the validation accuracy is 97.49, 96.14, and 96.07. The second model has a faster prediction time and a higher training and validation accuracy for all three sets. The training accuracy is 98.37, 98.08, and 97.86 and the validation accuracy is 98.55, 97.40, and 98.31.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Biometric fingerprint identification in low-quality samples using a hybrid features extraction method with deep learning

  • 期刊论文

  • 发表日期：

    2025年7月28日

查看全文
  • Alyaa Haleem,
  • Saoussen Ben Jabra,
  • Walid Chainbi

摘要

Fingerprint recognition is essential element of biometric security, yet it faces challenges related to accuracy, robustness, and latency. One significant issue is that the performance of fingerprint authentication systems significantly decreases when faced with low-quality fingerprints. In this paper, two new hybrid approaches are suggested to enhance the accuracy of fingerprint verification and create a reliable identification system. The first approach, SIFT-Descriptor-SCNN, combines the scale-invariant feature transform (SIFT) method with the proposed Siamese convolutional neural network model. First, we compute the SIFT feature descriptors and then pass them to the SCNN for further feature extraction and matching. This increases the system’s ability to distinguish between various patterns for accurate identification. The second novel approach combines oriented FAST with a dense grid, the SIFT algorithm, and the SCNN, named oriented FAST-Dense-SIFT-Descriptor-SCNN. In this work oriented FAST with a dense method is implemented for the detection of keypoints and then the detected keypoints are fed into SIFT algorithm to compute the descriptors. The integration of these algorithms with the SCNN model enhances the performance of the fingerprint identification system. We test the performance of the two hybridization models, the SOCOFing dataset, which includes 6000 real fingerprint and 49,270 synthetically altered versions organized into easy, medium and hard was used. The first model provides an accuracy of 97.57, 96.72, and 95.61 for the easy, medium, and hard datasets, while the validation accuracy is 97.49, 96.14, and 96.07. The second model has a faster prediction time and a higher training and validation accuracy for all three sets. The training accuracy is 98.37, 98.08, and 97.86 and the validation accuracy is 98.55, 97.40, and 98.31.