Training set optimization in 3D human face recognition by RBF neural networks

In the Neural Networks approach by Radial Basis Function RBF, the property of interpolation between faces, their variation, and the diversity of faces helps to minimize the output error. However, the training set size has to be optimized because the time to train an Artificial Neural Network ANN, is correlated with the number of samples in that training set. In this case, the samples are represented by 3D faces, where each point of this height matrix is important for the recognition task. Therefore, the amount of information used to identify a face is large. A test was performed to reach the best ratio between solution error and processing time, finding the minimal training set required for a specific solution error.