(For USM Staff/Student Only)

EngLib USM > Ω School of Electrical & Electronic Engineering >

Weighted sparse representation based classifier for finger vein recognition

Weighted sparse representation based classifier for finger vein recognition / Seow Wei Wen
Dalam tesis ini, klasifikasi berdasarkan perwakilan jarang wajaran (WSRC) direka bentuk untuk kegunaan pengecaman urat jari. Klasifikasi ini direka bentuk untuk meningkatkan prestasi konvensional klasifikasi berdasarkan perwakilan jarang (SRC) yang telah digunakan dalam salah satu penyelidikan untuk pengecaman urat jari. Salah satu kertas penyelidikan untuk pengecaman muka mencadangkan WSRC yang menggunakan Gaussian kernel sebagai skim pemberat. Oleh sebab projek ini adalah mengenai pengecaman urat jari, penggunaan Gaussian kernel dijangka tidak meningkatkan prestasi pengelasan. Jadi dalam projek ini, beberapa jenis skim pemberat telah diuji termasuk Gaussian kernel untuk mengetahui jika WSRC masih lebih baik daripada SRC dalam aplikasi pengecaman urat jari. Selain itu, satu kaedah ambang telah digunakan untuk meningkatkan kelajuan WSRC. Eksperimen telah dijalankan untuk empat pangkalan data. Kadar pengiktirafan yang terbaik diperolehi dengan menggunakan WSRC menunjukkan peningkatan sebanyak 0.44%, 0.95%, 1.37% dan 0.74% masing-masing untuk pangkalan data FV-USM, HongKong, IDIAP dan SDUMLA-HMT. Skim pemberat terbaik untuk setiap pangkalan data masing-masing adalah Spearman correlation, 1.5powdis, Pearson correlation dan Heat kernel. Masa pengiraan selepas ambang telah dikurangkan sebanyak 11.91%, 20.37%, 16.79% dan 23.52% masing-masing untuk pangkalan data FV-USM, HongKong, IDIAP dan SDUMLA-HMT. Kesimpulannya, WSRC boleh mengelaskan lebih baik daripada SRC tetapi skim pemberat terbaik adalah berbeza bergantung kepada pangkalan data dan kaedah ambang boleh digunakan untuk mengurangkan masa klasifikasi untuk WSRC. _______________________________________________________________________________________________________ In this thesis, the weighted sparse representation based classifier (WSRC) is designed for the finger vein recognition purpose. This classifier is designed to improve the performance of conventional sparse representation based classifier (SRC) applied in one of the past finger vein recognition researches. One of the research paper on face recognition proposed the WSRC with the Gaussian kernel as its weightage scheme. Since project is about finger vein recognition, using Gaussian kernel might not improve the classification performance. So in this project, different type of weightage schemes were tested including the Gaussian kernel to determine if WSRC still remains better than SRC in finger vein recognition application. Besides that, a threshold method was applied to improve the speed of the WSRC. The experiments were conducted over four databases. The best recognition rate using WSRC showed an improvement by 0.44%, 0.95%, 1.37% and 0.74% for FV-USM, HongKong, IDIAP, and SDUMLA-HMT database respectively. The best weightage schemes for each database were Spearman correlation, 1.5powdis, Pearson correlation and Heat kernel respectively. The computational time after threshold was reduced by 11.91%, 20.37%, 16.79% and 23.52% for FV-USM, HongKong, IDIAP and SDUMLA-HMT database respectively. In conclusion, WSRC can perform better than SRC but the best distance metric differs depending on the database and threshold can be applied to decrease the classification time of WSRC.
Contributor(s):
Seow Wei Wen - Author
Primary Item Type:
Final Year Project
Identifiers:
Accession Number : 875006061
Language:
English
Subject Keywords:
weighted sparse representation based classifier (WSRC); designed; finger vein recognition
First presented to the public:
6/1/2016
Original Publication Date:
7/5/2018
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Electrical & Electronic Engineering
Citation:
Extents:
Number of Pages - 130
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2018-07-05 11:38:25.452
Date Last Updated
2019-01-07 11:24:32.9118
Submitter:
Mohd Jasnizam Mohd Salleh

All Versions

Thumbnail Name Version Created Date
Weighted sparse representation based classifier for finger vein recognition1 2018-07-05 11:38:25.452