Please use this identifier to cite or link to this item:
|Title:||Removal of redundant gabor features using genetic algorithm||Authors:||Dang, Ngoc Minh||Keywords:||DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Control engineering||Issue Date:||2010||Abstract:||Face data processing has been an active research fields for many decades due to its various applications in different areas such as entertainment system, smart cards, information security, biometric system, and surveillance system and law enforcement. Researchers tend to represent faces by feature-based approach thanks to its good performance. Amongst different types of features, Gabor features are widely used because they are robust against illumination and pose changes. Due to the limitation in computational power, the common practice is to down-sample the face image to reduce number of Gabor features generated. This practice limits the size of images that can be analyzed. Face image with high resolution, however, are still important in classification problems with high intra-group variation. As not all of the generated Gabor features are necessary, the main objective of this Final Year Project is to develop an efficient removal scheme of redundant features in order to employ images with large dimension in face data processing. In particular, Genetic Algorithm is used to provide a sub-optimal but fast selection of feature ensemble. In this project, a Mixture of Experts system is designed for face recognition problems. The base classifiers are trained by using Adaboost algorithm over the Gabor feature set extracted from each single Gabor filter. Genetic Algorithm is then applied to select the most discriminate ensemble of classifiers. Finally, the final result is obtained by fusing the final outputs of base classifiers in the selected ensemble. The aforementioned approach is implemented in family classification problem in which the intra-group variation is high. Face photos (resolution 80x95) of 10 families with different races and age groups are investigated. Three scenarios are tested: (1) identification of a family member, (2) identification of a missing child and (3) identification of a missing parent. This method also allows us to investigate the performance of Gabor wavelets with higher resolutions of frequency and orientation. From Matlab simulation, we find out that, using 40 Gabor filters from 80 Gabor filters of 8 frequencies x 10 orientations is better than using 40 Gabor filters of 5 frequencies x 8 orientations.||URI:||http://hdl.handle.net/10356/40894||Rights:||Nanyang Technological University||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Student Reports (FYP/IA/PA/PI)|
Page view(s) 20247
checked on Oct 19, 2020
checked on Oct 19, 2020
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.