AKHCRNet: Bengali Handwritten Character Recognition Using Deep Learning

  • 2020-09-15 16:12:54
  • Akash Roy
  • 0

Abstract

I propose a state of the art deep neural architectural solution forhandwritten character recognition for Bengali alphabets, compound characters aswell as numerical digits that achieves state-of-the-art accuracy 96.8% in just11 epochs. Similar work has been done before by Chatterjee, Dutta, et al. 2019but they achieved 96.12% accuracy in about 47 epochs. The deep neuralarchitecture used in that paper was fairly large considering the inclusion ofthe weights of the ResNet 50 model which is a 50 layer Residual Network. Thisproposed model achieves higher accuracy as compared to any previous work & in alittle number of epochs. ResNet50 is a good model trained on the ImageNetdataset, but I propose an HCR network that is trained from the scratch onBengali characters without the "Ensemble Learning" that can outperform previousarchitectures.

 

Quick Read (beta)

loading the full paper ...