BanglaNet: Bangla Handwritten Character Recognition Using Ensembling of Convolutional Neural Network

Handwritten character recognition is a crucial task because of its abundantapplications. The recognition task of Bangla handwritten characters isespecially challenging because of the cursive nature of Bangla characters andthe presence of compound characters with more than one way of writing. In thispaper, a classification model based on the ensembling of several ConvolutionalNeural Networks (CNN), namely, BanglaNet is proposed to classify Bangla basiccharacters, compound characters, numerals, and modifiers. Three differentmodels based on the idea of state-of-the-art CNN models like Inception, ResNet,and DenseNet have been trained with both augmented and non-augmented inputs.Finally, all these models are averaged or ensembled to get the finishing model.Rigorous experimentation on three benchmark Bangla handwritten charactersdatasets, namely, CMATERdb, BanglaLekha-Isolated, and Ekush has exhibitedsignificant recognition accuracies compared to some recent CNN-based research.The top-1 recognition accuracies obtained are 98.40the top-3 accuracies are 99.79BanglaLekha-Isolated, and Ekush datasets respectively.