Abstract:
Optical character recognition for Bangla handwritten character is an important task for
our daily life. Recognizing handwritten characters has difficulties because it differs from
person to person. Thus recognizing handwriting characters can be very challenging. In
this paper, I have introduced two different DCNN models for recognizing Bangla
compound characters.In first model, I have used 6 convolution layer, 3 dropout layer and
batch normalization in each layer with LeakyReLu activation function and other
proposed model has 6 layers of convolution layers and 2 dropout layers, ReLU as an
activation function to recognize 171 classes of Bangla handwritten character characters.
The model was tested on the AIBangla dataset of compound characters. My proposed
model was trained to recognize 171 characters using the DCNN model. Among those
two, model 2 provided highest accuracy of 76.12%. There is still room for improvement,
these results are significantly better than other models.
