Nepali Handwriting Recognition using Convolution Neural Network

The recognition of texts from the scanned image can have various applications, which are based on optical character recognition. This paper was proposed and carefully experimented to analysis and recognize handwritten Nepali character using Convolution Neural Network. The preliminary experiment has been done with 92 thousand images of 46 different classes of 32 * 32 characters of Nepali Handwriting which went through different preprocessing stages like clipping and cropping, grayscale conversion and through different processes like feature extraction etc. The recognition has been experimented with the help of template matching technique. This proposal will employ Back Propagation algorithm along with Gradient descent algorithm will be used to update the weights, an artificial neural network training and testing. Thus, this experiment concluded that the convolution neural network model has more accuracy than the Feed Forward neural network in character recognition.