Classification of Characters Using Multilayer Perceptron and Simplified Fuzzy ARTMAP Neural Networks

There are various types of methods that can be used to recognize and classify the targeted object in the field of pattern recognition. Thus, this paper presents the classification of characters by combining the features based on Moment Invariant (MI) and Artificial Neural Network (ANN). The moment invariant is used to extract the feature image based on translation, scaling and rotation (RTS) independently in order to test the invariant properties. In this study, the type of moment invariant that has been used is Geometric Moment Invariant (GMI). This moment invariant will produce seven feature vectors which will later be used as the input features for the classification process. In addition, the current study has also utilized the potential of ANN in order to classify the image based on its category. Here, there are two types of ANN that are used to recognize the character image which are Multilayer Perceptron (MLP) and Simplified Fuzzy ARTMAP (SFAM) neural networks. To train the MLP network, the algorithm of Levenberg-Marquardt is adopted in order to check the applicability. Based on the classification that has been computed, the results show that both networks have produced good classification performance with overall accuracy above 90%. However, the MLP trained by Levenberg-Marquardt (MLP_LM) shows the highest classification performance with 94.46% as compared to the SFAM network.