Open Access Open Access  Restricted Access Subscription or Fee Access

Bangla Handwritten Digit and Basic Letter Recognition Using Machine Learning Techniques

Juthika Majumder, Razia Sultana, Rameswar Debnath

Abstract


Abstract

The usage of computer is increasing day by day in Bangladesh, so the use of Bangla is increasing in computer. Furthermore, the use of Bangla handwritten character is also increasing in many computer applications. There exist many techniques for recognition of handwritten character. From the studies we see that machine learning techniques are better techniques for recognition of handwritten character than others. A necessary prerequisite for measuring the performance of machine learning techniques is a large dataset for training. In this paper our contributions are two-fold. The first one is to develop large real dataset. We have developed a dataset containing 25,000 samples of Bangla handwritten digits and a dataset containing 10,200 samples of Bangla handwritten letters. Each sample in the databases is an image of size 16´16 pixels. The second contribution is to a systematic evaluation of the performance of machine learning techniques for Bangla character recognition. In this paper, we choose mostly used machine learning techniques: one-against-one support vector machine, one-against-all support vector machine, artificial neural networks, and radial basis neural network. We apply several morphological operations on image dataset that can improve the performance of machine learning techniques for handwritten character recognition. From the experimental results we see that classifiers show better results with preprocessed data, and support vector machines give significantly better recognition results than those of radial basis neural network and artificial neural networks. From the results we see that 99.10% accuracy is obtained for digit recognition and 96.08% accuracy is obtained for letter recognition by support vector machine.

 

Keywords:Support vector machine, artificial neural network, radial basic function network, principal component analysis, flood fill algorithm, thinning, dilation, opening, binarization

Cite this Article

Juthika Majumder, Razia Sultana, Rameswar Debnath. Bangla Handwritten Digit and Basic Letter Recognition Using Machine Learning Techniques. Journal of Image Processing & Pattern Recognition Progress. 2016; 3(3): 1–15p.



Full Text:

PDF

Refbacks

  • There are currently no refbacks.