Recognition of bangla basic characters using multiple classifiers

Abstract

Character recognition is an important area in image processing and pattern recognition fields. A novel scheme for recognition of offline basic characters of Bangla using multiple classifiers is described here. Compared to English characters, there are different complex shaped characters in Bangla alphabet. Dealing with such a large number of characters with a suitably designed feature set is a challenging problem. Moreover, such a large variety of complex shaped characters, some of which have close resemblance, make the problem more difficult. Considering the complexity of the problem, present approach makes an attempt to identify the basic characters. We have adopted this hybrid approach because it is nearly impossible to find a set of stroke features which are sufficient to classify the characters. A prototype of the system is tested with a data set containing 4423 characters of different font and size. On average, the recognition accuracies for Binary tree based classifier and Multilayer perceptron [with backpropagation for learning] (MLP) are 90% above approximately.