Non-Conventional Approaches to Feature Extraction for Face Recognition
Jozef Ban, Matej Féder · 2011
This paper deals with human face recognition based on the use of neural networks, such as MLP (multi-layer perceptron), RBF (radial basis function) network, and SVM (support vector machine) methods. The methods are tested on the MIT (Massachusetts Institute of Technology) face database. We use non-conventional methods of feature extraction for the MIT face images in order to improve the recognition results. These methods use so called HLO and INDEX images. HLO images are generated by feature extraction of the MLP neural network in auto-association mode, and INDEX images are formed by a self-organized map used for image vector quantization. We propose novel methods based on HLO and INDEX images with SVM classifier. We also analyze the impact of adding noise to the learning process.