Offline handwriting recognition on hindi using CNN RNN hybrid network

Abstract

Recognising the Scripts written in Hindi is an onerous task due to the delicacies and variations in the script. Handwriting of every script is diversified in writing styles and orientation angles. This requires a neural network with a large training dataset to successfully recognise the script. Till now the development of this handwriting recognisers have been obstructed due to the unavailability of hand-written public datasets in Hindi. In this paper we used CNN RNN hybrid network based on IIIT-HW-Dev dataset to differentiate and recognise handwritten scripts.