Bangla Handwritten Character Recognition: an overview of the state of the art classification algorithm with new dataset

Recognition of handwritten characters from Bangla handwritten texts is of immense importance considering the complexity of the task. Researchers have explored the task of recognizing Bangla handwritten digits, but a few numbers of published works are available for Bangla Handwritten Character Recognition (BHCR). In our paper, we present a comparative overview of classification algorithms for BHCR, which may help the researcher to decide an appropriate classification algorithm for their work. We have created a new dataset of Bangla handwritten characters from 150 volunteers at different levels. We extracted around 2500 samples of Bangla characters, which consist of Bangla Vowels only. Histogram adjustment and other image preprocessing techniques are applied in handwritten characters before their classification. We compare the performance of seven commonly used classification algorithms for BHCR in terms of Sensitivity, Miss Rate, Specificity, Precision, Fall-out, F-score, and Overall Accuracy. This result shows that among the seven algorithms, ANN (Artificial Neural Network) performed best. LR (Logistic Regression) performed well compared to others in terms of the standard measures like sensitivity, specificity and error rate. This comparative overview will help scientists, especially the new researchers to give a quick start with Bangla handwritten character recognition.