Printed digits recognition using multiple multi Layer perceptron classifiers and Hu moments
Khedidja Derdour, Leïla Hayet Mouss · 2015
This paper presents the combination of MultiLayer Perceptron (MLP) artificial neural network classifier for printed Arabic digits recognition. Different types of feature are used, cavity, zoning, pixel feature and Hu moment invariants which are invariant under change of size, translation, and orientation (rotation). On experimentation with a database of 6240 samples in multi-font and multi-size. We propose an approach based on a combination of Multi-Layer Perceptron using different feature types. The technique yields an average recognition rate of 94.33% evaluated after three methods of learning Kfold cross validation holdout and resubstitution method. The proposed printed digit recognition system can be used within applications related to post office (postal address, postal sorting), car plates, barcode and can also be extended to recognize handwritten Arabic digit. Keywords— combination of classifiers , Hu moments, multi layer perceptron, artifical neural networks, printed digit recognition, feature extraction.,