000			04555ntm a2200373 i 4500
999			_c91085 _d91091
003			MY-KuUP
005			20251125105420.0
006			t||||fr|||| 000 0
007			ta
008			191107t20182018my a|||fram|| 000 0 eng d
020			_aTHE0008400(Local) _qhardback
040			_aUMP _beng _cUMP _erda
090			_aFKEE .T56 2018 r Thesis
100	1		_aTing, Ei Wei, _eauthor.
245	1	0	_aDevelopment of finger vein identification system using feature combination and dimensionality reduction / _cTing Ei Wei
264		1	_aKuantan, Pahang : _bUMP, _c2018
264		4	_c© 2018
300			_axiii, 76 pages : _billustrations ; _c30 cm. + _e1 CD-ROM
336			_atext _2rdacontent
336			_atext _2rdacontent
337			_aunmediated _2rdamedia
337			_acomputer _2rdamedia
338			_avolume _2rdacarrier
338			_acomputer disc _2rdacarrier
347			_atext file _bPDF _2rda
500			_aFaculty of Electrical and Electronics Engineering
502			_aThesis (Master of Science) -- Universiti Malaysia Pahang – 2018
504			_aIncludes bibliographical references
520	3		_aWith the increase in globalization and living standard, finger vein biometric trait becomes a popular focus in the human recognition and security system. Finger vein is high security as it is hard to be stolen or duplicate since it hides underneath the finger skin. Image quality and contrast is the main problem that was faced by researcher in the finger vein identification system. It may cause by the wrong posture, light intensity and image rotation during the image captured session. Previous works had stated that the wavelet-based features are invariant to image rotation while local binary based features robust against high saturation and irregular shading. However, the feature combination of these two different features type has not been studied yet in finger vein identification system. Besides that, the dimensionality reduction technique such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) was frequently involved in finger vein recognition system in this recent year. Generally, PCA was used to remove the noise residing in the discarded dimension. However, the features reduction in earlier stage using PCA may remove the important features as PCA might remove the local information where it concludes that the information for the corresponding eigenvalue is noise. Therefore, this research proposed to develop a finger vein identification using combination features of Discrete Wavelet Transform (DWT) and Local Binary Pattern (LBP). The combined features were then undergoing dimensionality reduction using PCA and LDA. The finger vein image firstly undergoes image pre-processing to enhance the image quality, followed by region of interest (ROI) extraction and noise removal using median filtering. The pre-processed finger vein images will go through feature extraction using DWT and LBP to get the features in vectors form and then combined. The process was continued by applying PCA or LDA on the combination features. Linear Support Vector Machine (SVM) and Fine K-Nearest Neighbours (FKNN) were used for classification in this research. At the same time, the conventional finger vein identification system was designed with Repeated Line Tracking (RLT) and Maximum Curvature (MAXC) as feature extraction methods followed by Speeded-up Robust Features (SURF) as the classification method. The performance of the proposed system was compared with the conventional finger vein identification system using same database which is known as SDMULA-HMT. Throughout the experiments, the feature combination of DWT and LBP was proved to perform better than the finger vein identification system that involved only single type of feature by average of 6.91% improvement. At the early stage, the proposed system achieved 80.16% identification rate, which is a bit lower than the conventional system. However, when the proposed system and the conventional system are applied with dimensionality reduction using PCA and LDA, the identification rate increased dramatically in proposed system while there was only small improvement on the conventional finger vein identification system. The proposed system achieved the highest identification rate at 100% when the feature is performed using LDA at feature dimension of 30.
610	2	0	_aFaculty of Electrical and Electronics Engineering _xDissertations
650		0	_aUniversities and colleges _xDisertations
650		0	_aTheses
942			_2lcc _cTHESIS