| Summary: | Today, finger vein has become a new biometric technology. The challenge of
finger vein recognition comes to during the process of feature extraction. Since the
low contrast finger vein images may contain shading and noise, it is important to
precisely preprocess, extract and preserve the vein patterns. Algorithms such as
Gabor Filter, Local Line Binary Pattern (LLBP) and Principal Component Analysis
(PCA) have been proposed in recent study to extract finger vein features. In this
thesis, a modified pattern map feature extraction method is proposed for finger vein
recognition. Instead of obtaining fmger vein features from multi-filtered images, the
features images are generated from pattern templates which are the eigenveins
obtained from PCA process. Every fmger vein image is then transformed into
pattern map images from a pattern matching process between an input finger vein
image and the pattern templates. Finally, nearest neighbour classifier with Euclidean
distance metrics is used for classification. The main contribution of this thesis is the
new way of generating pattern templates, which selects small blocks from every
class within an area of constraint. Additionally, pattern map is implemented in finger
vein recognition for the first time. Experimental results show that the proposed
pattern map algorithm has better performance with a consistent identification rate of
99% and above compared to existing methods such as PCA and Gabor with FVCode.
This shows that pattern map is a reliable feature extraction method and is able to
represent finger vein pattern effectively.
|
|---|