Loading...

Palm Vein Pattern Recognition using Deep Convolutional Neural Network (DCNN) with Gabor Filter

Nazari Tavakoli, Amir Ali | 2023

66 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56385 (52)
  4. University: Sharif University of Technology, International Campus, Kish Island
  5. Department: Science and Engineering
  6. Advisor(s): Motahari, Abolfazl; Peyvandi, Hossein
  7. Abstract:
  8. Frequently using Personal Identification Information has escalated the security concerns of bank accounts, emails, daily transactions, and other activities. Therefore, user access to such apps must be controlled. Traditional personal verification methods offer limited security because they might need to be remembered or stolen. Therefore, Biometric authentication, which identifies persons by their unique biological information, is gaining popularity. However, palm vein identification is highly secure because the vein patterns are not duplicated in other people, even in monozygotic twins. Moreover, it has a liveness detection and is convenient since the vein pattern cannot be faked, forgotten, or stolen. In this thesis, we consider using the Tongji University Contactless Palm vein dataset to study the problem of palm vein recognition. Images were collected from 300 participants for use in their dataset. It has two different sessions. During each session, the participant was requested to produce ten photos of each palm. There are 12,000 photos in the database, all taken from one of 600 distinct palms. The most extended observed interval was 106 days, while the shortest period was 21 days. We intend a DCNN-based architecture with the Inception-ResNet-v1 model and the Gabor filter in the first convolution layer of the stem structure to improve the result, and we use it as a feature extractor for the identification or verification phase of the process. It is also faster than the simple DCNN model in training speed. We demonstrated that our technique could correctly identify all images by measuring the recognition accuracy and the precision-recall rate, which means that our method can identify all photos correctly. For verification, we look at the Euclidean distance between the feature vectors of the two palms evaluated to see how well they match. Our approach has been proven to achieve a better EER of 2.2% compared to other methods
  9. Keywords:
  10. Deep Convolutional Neural Networks ; Biometrics ; Gabor Filters ; Pattern Recognition ; Palm Vein Recognition ; Biometric Authentication

 Digital Object List

 Bookmark

No TOC