000			04564nam a22003737a 4500
999			_c94832 _d94838
003			MY-KuUP
005			20251125105810.0
006			a||||fr|||| 000 0
007			ta
008			201113t20202020my a|||frma|| 000 0 eng d
020			_aTHE0008583(Local)
040			_aUMP _beng _cUMP _erda
090			_aFKOM .S26 2020 r Thesis
100	1		_aSam Yin Yee, _eauthor.
245	1	0	_aDynamic fast local laplacian completed local ternary pattern (dynamic flapcltp) for face recognition / _cSam Yin Yee
264		1	_aKuantan, Pahang : _bUMP, _c2020
264		4	_a© 2020
300			_axvii, 129 pages : _billustrations (some color) ; _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 Computing
502			_aThesis (Master of Science) -- Universiti Malaysia Pahang – 2020
504			_aIncludes bibliographical references
520	3		_aToday, face recognition has become one of the typical biometric authentication systems used for high security. Some systems may use face recognition to enhance their security and provide high protection level. Feature extraction is considered to be one of the most important steps in face recognition systems. The important and interesting parts of the image in feature extraction are represented as a compact feature vector. Many features, such as texture, colour and shape, have been proposed in the image processing fields. These features can also be classified globally or locally depending on the image extraction area. Texture descriptors have recently played a crucial role as local descriptors. Different types of texture descriptors, such as local binary pattern (LBP), local ternary pattern (LTP), completed local binary pattern (CLBP) and completed local ternary pattern (CLTP), have been proposed and utilised for face recognition tasks. All these texture features have achieved good performance in terms of recognition accuracy. Although the LBP performed well in different tasks, it has two limitations. LBP is sensitive to noise and occasionally fails to clearly distinguish between two different texture patterns with the same LBP encoding code. Most of the texture descriptors inherited these limitations from LBP. CLTP is proposed to overcome the limitations of LBP. CLTP performed well with different image processing tasks, such as image classification and face recognition. However, CLTP suffers from two limitations that may affect its performance in these tasks: the fixed value of the threshold value that is used during the CLTP extraction process regardless of the type of dataset or system and the longer length of the CLTP histogram than that of previous descriptors. This study focused on handling the first limitation, which is the threshold selection. Firstly, a new texture descriptor is proposed by integrating the fast-local Laplacian filter and the CLTP descriptor, namely, fast-local Laplacian CLTP (FLapCLTP). The fast-local Laplacian filter can help in increasing the performance of the CLTP due to its extensive detail enhancements and tone mapping; this contribution is handled by the constant threshold value used in CLTP. A dynamic FLapCLTP is then proposed to address the aforementioned issue. Instead of using a fixed threshold value with all datasets, a dynamic value is selected based on the image pixel values. Therefore, each different texture pattern has different threshold values to extract FLapCLTP from the pattern. This dynamic value is automatically selected according to the centre value of the texture pattern. Therefore, a dynamic FLapCLTP is proposed in this study. Finally, the proposed FLapCLTP and dynamic FLapCLTP are evaluated for facial recognition systems using ORL Faces, Sheffield Face, Collection Facial Images, Georgia Tech Face, Caltech Pedestrian Faces 1999, JAFFE, FEI Face and YALE datasets. The results showed the priority of the proposed texture compared with previous texture descriptors. The dynamic FLapCLTP achieved the highest recognition accuracy rates with values of 100%, 99.96%, 99.75%, 99.69%, 94.86%, 90.33%, 86.86% and 82.43% using UMIST, Collection Facial Images, JAFFE, ORL, Georgia Tech, YALE, Caltech 1999 and FEI datasets, respectively.
610	2	0	_aFaculty of Computing _xDissertations
650		0	_aUniversities and colleges _xDissertations
650		0	_aTheses
942			_2lcc _cTHESIS