Wong , Edward Kie Yih (2009) Design of a biometric authentication methodology using palmprint. Masters thesis, Universiti Malaysia Sabah.
Biometric is the science of measuring human characteristics for the purpose of authentication or identification. Palmprint, one of the human physiological characteristics, is gaining attention among researchers as the mean of security. This is because palmprint is rich in unique features. The objectives of this thesis are to design an image processing methodology for palmprint image extraction, to design an efficient feature extraction method and an efficient classification method for palmprint biometric. The hand image is acquired using a digital camera. From the acquired hand image, the hand region are segmented from the background. The gaps between the fingers are located and marked as key pOints. Using the key pOints, the palmprint image rotational angle and the palmprint area are estimated. The palmprint image is rotated and extracted from the hand image. The palmprint image is enhanced using image adjustment and histogram equalization. The enhanced palmprint image is resized to a predefined size. Five types of palmprint features, namely, Geometric Line features, Discrete Cosine Energy features, Wavelet Energy features, Sequential Haar Energy features and SobelCode features, are extracted from the palmprint image. The palmprint features are represented using different feature representation method to form feature vectors. The feature vector is compared with the palmprint database using Hamming Distance similarity measurement, Euclidean Distance similarity measurement and Scaled Conjugated Gradient (SCG) based feedforward backpropagation neural network. Different types of feature extraction methods and classification methods have been tested to find the best discriminating palmprint feature. For comparison using distance similarity, the SobelCode features can achieve up to 97.44% accuracy, followed by the Discrete Cosine Energy features with 94.58%. The Sequential Haar Energy feature and Wavelet Energy feature can achieve 93.99% and 92.95% accuracy respectively. Using SCG-based feedforward backpropagation neural network, the accuracy of the Wavelet Energy Feature and Discrete Cosine Energy Feature can be slightly increased to 99%. Geometric Line Feature can achieve 89.41% accuracy.
|Item Type:||Thesis (Masters)|
|Uncontrolled Keywords:||design, biometric, methodology, palmprint, security, human characteristic, image extraction|
|Subjects:||Q Science > QK Botany|
|Divisions:||SCHOOL > School of Engineering and Information Technology|
|Deposited By:||IR Admin|
|Deposited On:||10 Jun 2013 15:12|
|Last Modified:||10 Jun 2013 15:12|
Repository Staff Only: item control page