Face Recognition by Global Optimal Discriminant Features and Ensemble Artificial Neural Networks
Jingjing Wang, Jianqin Yin · 2009
Good feature extraction scheme and classifiers are the key to face recognition algorithms. A general and efficient face feature extraction approach is presented which utilizes linear discriminant information and global search strategy. In order to get rid of redundant information and meanwhile reduce computational burden, we first compute the nonzero feature space of scatter matrix of the training set, and then perform a global search on it to seek out the most valuable discriminant information of faces. Genetic algorithm is used for searching because of its well-known global search ability. Also, good classifiers are designed by using ensemble artificial neural networks. Based on the individual classifiers, appropriate combinations are selected to construct the classification committee using EDAs (estimation of distribution algorithms). Experiments are designed to test the performance of the feature extraction method and the classifiers separately. Results show that the proposed method produces good recognition rates on three benchmark databases.