![]() | Only 14 pages are availabe for public view |
Abstract A face recognition system based on rough set theory and neural networks is developed in this thesis. The system consists of three stages; principal component analysis, rough set theory for dimensionality reduction, and neural network for recognition. In this thesis use peA for feature projection and reduction, peA has an apparent limitation it cannot guarantee that a few first selected principal components are the most adequate features for face recognition. To solution this limitation is apply Rough set theory for feature selection. Applying the rough set theory to select the most adequate and discriminative features from the principal components generated by peA then using LVQ neural network for classification. In this work, a single network was build for all person. the number of neurons in the output neural network equal to the number of people to be classified. When a person is recognized, the neuron corresponding in the last layer will have a value 1, and the other neurons will have a value O. Using peA and rough sets for feature extraction and selection, then using LVQ NN for classification shown significant improvement in case of training time requirement i.e. the classifier network converges faster and the recognition rate has also increased. The algorithms that have been developed are tested on ORL and Yale Face Databases. |