Fusion of holistic and local features for palmprint recognition
Palmprint recognition has become an important and rapidly developing technology in biometric system over the past decade. The success of palmprint identification requires the best matching of the test sample from input data and the templates in the palmprint database. The used of holistic and loc...
Saved in:
| Format: | Thesis |
|---|---|
| Language: | English |
| Subjects: | |
| Online Access: | http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/76738/1/Page%201-24.pdf http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/76738/2/Full%20text.pdf http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/76738/3/Declaration%20Form.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Palmprint recognition has become an important and rapidly developing technology in
biometric system over the past decade. The success of palmprint identification requires
the best matching of the test sample from input data and the templates in the palmprint
database. The used of holistic and local features separately will have limitations in
geometry and variations. Information fusion of palmprint biometric is developed in
order to produce a better recognition performance. Thus, this research work presented
the fusion of holistic and local features for palmprint recognition. The overall structure
of the study takes the form of three major steps includes pre-processing techniques,
feature extraction and matching process before proceeds with fusion to combine the
holistic and local features. Pre-processing technique is the initial stage to make sure the
palmprint image from dataset is cropped and resized into the specific size. The entire
proposed method is validated using benchmark PolyU dataset for palmprint recognition
analysis. Gabor filter pattern is used to extract important information in holistic features
while Discrete Cosine Transform (DCT) is used to extract low frequency energy of
local features. Then, the computation of Principal Component Analysis (PCA) is
applied to reduce the high dimensional feature space to low dimensional feature space.
Low dimensional feature space preserved low frequency information. Classification is
the process used to distinguish and classify a new observation based on the training set
of data. The purpose of Euclidean distance classifier is to measure the matching value
and its closeness between the training and testing feature vectors. Among various fusion
levels, matching score level fusion is the most suitable approach in combining the
match score from two different matchers because this fusion rule can increase matching
accuracy. It is developed by forming a single value for decision process from the
matching output of different matching module. In the matching score fusion scheme,
weighted sum rule produced superior performance. The best recognition rate of 97% is
achieved using 100 subjects. Based on the recognition analysis, there are three
important parameters that affect the performance which is the size of input image, the
effect of principal components number, and the number of DCT coefficient. |
|---|