A Novel Multilayer Perceptron Artificial Neural Network based Recognition for Kurdish Manuscript

Behnam Zebardast

doi:10.17485/ijst/2014/v7i3.3

Abstract

Alphabetic recognition is one of the most interesting and successful research fields in artificial intelligence and pattern recognition. The different writing structures of the languages and the presence of different approaches in diagnosis of letters of the different languages has been a challenge in alphabetic recognition. All these challenges have made many researches switch this research area. Of the different approaches in alphabetic recognition, the neural artificial networks could have been successful according to the capability of parallel process and learning capability for a special application like recognition pattern. So, it is a suitable approach in alphabet recognition. The Kurdish language holds two manuscripts according to Latin and Arabic alphabets. In this paper the Multilayer Perceptron (MLP) artificial neural networks are studied by the back-propagation algorithm to recognize the Kurdish-Latin manuscript. The proposed method is also applicative for diagnosis of all letters of the other Latin languages like English, Italian and etc. In this paper the MLP artificial neural networks are implemented in MATLAB environment. The efficiency factor for recognition of the Kurdish letters is to maximize the recognition accuracy of the Kurdish letters in training and testing the MLP artificial neural networks. This accuracy is 85.1535% in training stage and 81.2677% in testing stage.

Highlights

Many important steps have been taken for identifying the pattern in late decades
In this paper we have introduced the related works in the field of recognition of the English, Arabic, Persian and Indian languages letters with different models of artificial neural networks and have tried to make the researchers and the readers of this paper clearly understand the letters diagnosis of the different languages using artificial neural networks models so it is not possible to compare the works about the none Latin- Kurdish letters recognition to the results of this paper which is about the recognition of the Kurdish letters via Multilayer Perceptron (MLP) artificial neural networks
Researchers[18] have used MLP artificial neural networks using the back-propagation algorithm and the Radial Basis Function (RBF) artificial neural networks for recognition of the Hindi letters. In this method a dataset which includes 245 repetitive characters of Hindi letters is used in which 125 characters are used as training data and 120 characters are used as testing data. In this method the results showed that the MLP artificial neural networks using learning back-propagation algorithm was better in recognition accuracy and memory usage

Summary

Introduction

Many important steps have been taken for identifying the pattern in late decades. A pattern can be a letter or a sample manuscript, a goal for radar, speaking signal, cardiograph signal, signature, fingerprint or even a part of the parts of production line in a factory[1,2]. There are diverse kinds of languages which leads to the challenges in letter diagnosis, the automation process of the pattern recognition machines has made this easier[2, 3]. A Novel Multilayer Perceptron Artificial Neural Network based Recognition for Kurdish Manuscript been proposed because the artificial neural networks can fulfill some shortages of the common pattern recognition methods and represent interesting results[6]. The proposed method in this paper is used for the letter recognition of all other Latin languages like English, Italian and etc.

Related Works

Recognition of the Kurdish Language Letters

Proposed Model

Findings

Conclusions and the Future