Register Transfer Level Implementation Of Pooling - Based Feature Extraction For Finger Vein Identification
Recently, finger vein biometric identification methods have had more attention among the researchers due to its various advantages such as: uniqueness to individuals, immunity to ages and invisibility to human eye (hard to duplicate). Many improvements methods were utilized to increase the speed and...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2017
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/39748/1/Mohamed_Hassan_Mohamed_Ali_24_Pages.pdf http://eprints.usm.my/39748/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Recently, finger vein biometric identification methods have had more attention among the researchers due to its various advantages such as: uniqueness to individuals, immunity to ages and invisibility to human eye (hard to duplicate). Many improvements methods were utilized to increase the speed and accuracy of the identification. Feature extraction techniques based on global feature extraction such as Principle Component Analysis (PCA) were implemented. However, the results did not show robustness to occlusions and misalignments on the finger vein images. Therefore, local feature extraction techniques were used to overcome these issues. A pooling based feature extraction technique for finger vein identification was implemented in this research. The proposed algorithm extracted the local feature information of the finger vein pattern (patches), and used these patches to improve the robustness of the identification. The algorithm was mainly inspired by spatial pyramid pooling in generic image classification combined with PCA. With patch size = 4, four pyramid levels = [1x1, 2x2, 3x3, 4x4] and ~38 % dimension reduction on the extracted features vector (10 PCA coefficient), the accuracy of the identification was 88.69 % which was higher than PCA by 10.10%. The proposed algorithm was implemented on hardware using Verilog-HDL, and targeting Field Programmable Gate Array (FPGA) applications. The result showed an outstanding speed improvement compared to software implementation. The time consumed by the hardware for extracting the features of one image was 310X time faster than the consumed time for software implementation. With those improvements in accuracy and the speed, the proposed algorithm contributes to the advancement of finger vein biometric system. |
|---|