A modular neural system for handwritten alphabet recognition using invariant moment-based features

A recognition system for identifying handwritten alphabets using a specific artificial neural network structure is described. Layers of feedforward networks preprocess the alphabets by extracting features that enhance the characters´ dissimilarity. The features approximate the moments and moment invariants which are invariant to translation, rotation, and scaling of the processed alphabets. The collective tasks of the feedforward network modules integrate feature extraction preprocessing with classification