Please use this identifier to cite or link to this item:
Title: Weighted voting component-based face recognition
Authors: Bazzaz, Hussein Al
Advisors: Dagher, Issam 
Subjects: Human face recognition (Computer science)
Issue Date: 2017
The face Recognition researched within this paper compares the component based on the block and global based recognition methods. The method presented in this paper consists of the following major steps: 1- Facial component detection and cropping using one of the following: a- The Viola-Jones Object Detection Framework b- Facial Landmarks 2 - Component Representation Block and component-based face recognition methods rely on the histogram of oriented Gradients (HOG features) to represent the facial components, while the global-based face recognition uses the distance vector to represent the face. 3 - Recognition Model Training Block and component-based face recognition methods train a separate model for each facial component. The global based face recognition method trains one model for each person. KNN and SVM machine learning methods have been used and compared in each case. Three public databases were used: 1 - AT&T with 40 subjects and 400 images. 2 - PUT Database with 50 subjects and 1100 images. 3 - AR database with 50 subjects and 1300 images. The Improvement observed using the method presented in this paper are as such: 1- nearly 100% accurate in detecting facial components during all pose changing circumstances. 2- Improve the machine learning classification accuracy by combining multiple Classifications Through the method of majority voting, allowing the ability to overcome the illumination and pose change at the same time.
Includes bibliographical references (p. 71-77).

Supervised by Dr. Issam Dagher.
Rights: This object is protected by copyright, and is made available here for research and educational purposes. Permission to reuse, publish, or reproduce the object beyond the personal and educational use exceptions must be obtained from the copyright holder
Ezproxy URL: Link to full text
Type: Thesis
Appears in Collections:UOB Theses and Projects

Show full item record

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.