3.2. Neural Classifier Configuration
The classifiers chosen for the task of SPV and character classification was a feed-forward Multi-layered Perceptron (MLP) trained with the resilient backpropagation (BP) algorithm. For experimental purposes, the architectures were
modified varying the number of inputs, outputs and hidden units. The number of inputs to each network was associated with the size of the feature vector for each image. Various vector dimensions were investigated. The most uccessful
vector configurations were of size 80 for SPV and 120 for character classification (using MDF).