Throughout recent years Machine Learning has acquired attention, due to the abundant data. Thus, devising techniques to reduce the dimensionality of data has been on going. Object detection is one of the Machine Learning techniques which suffer from this draw back. As an example, one of the most famous object detection frameworks is the Viola-Jones Rapid Object Detector, which suffers from a lengthy training process due to the vast search space, which can reach more than 160,000 features for a 24X24 image. The Viola-Jones Rapid Object Detector also uses Adaboost, which is a brute force method, and is required to pass by the set of all possible features in order to train the classifiers. Consequently, ways for reducing the whole feature set into a smaller representative one, eliminating those features that have non relevant information, were devised. The most commonly used technique for this is Feature Selection with its three categories: Filters, Wrappers and Embedded. Feature Selection has proven its success in providing fast and accurate classifiers. Wrapper methods harvest the power of evolutionary computing, most commonly Genetic Algorithms, in finding the set of representative features. This is mostly due to the Advantage of Genetic Algorithms and their power in finding adequate solutions more efficiently. In this thesis we propose GAdaboost: A Genetic Algorithm to accelerate the training procedure of the Viola-Jones Rapid Object Detector through Feature Selection. Specifically, we propose to limit the Adaboost search within a sub-set of the huge feature space, while evolving this subset following a Genetic Algorithm. Experiments demonstrate that our proposed GAdaboost is up to 3.7 times faster than Adaboost. We also demonstrate that the price of this speedup is a mere decrease (3%, 4%) in detection accuracy when tested on FDDB benchmark face detection set, and Caltech Web Faces respectively
Computer Science & Engineering Department
MS in Computer Science
Committee Member 1
Committee Member 2
The author retains all rights with regard to copyright. The author certifies that written permission from the owner(s) of third-party copyrighted matter included in the thesis, dissertation, paper, or record of study has been obtained. The author further certifies that IRB approval has been obtained for this thesis, or that IRB approval is not necessary for this thesis. Insofar as this thesis, dissertation, paper, or record of study is an educational record as defined in the Family Educational Rights and Privacy Act (FERPA) (20 USC 1232g), the author has granted consent to disclosure of it to anyone who requests a copy.
Institutional Review Board (IRB) Approval
Not necessary for this item
(2016).GAdaboost: Accelerating adaboost feature selection with genetic algorithms [Master's Thesis, the American University in Cairo]. AUC Knowledge Fountain.
Tolba, Mai. GAdaboost: Accelerating adaboost feature selection with genetic algorithms. 2016. American University in Cairo, Master's Thesis. AUC Knowledge Fountain.