Offline Handwritten Gurumukhi Character Recognition System Using Deep Learning

Currently, Gurumukhi-a religion-specific language originating from India, ranks as the 14th most spoken language and the 18th most popular writing script language of the entire world. However, while there exists a large body of literature related to recognition of handwritten texts in various languages, number of publications related to recognition of Indian handwritten scripts is considerably smaller. It concerns also the case of Gurumukhi. Hence, in the current contribution, we consider Gurumukhi handwritten character recognition, to fill the existing practical gap. The proposed approach is based on deep convolutional neural networks and has been applied to the 35 core Gurumukhi characters. Obtained results are promising, as accuracy of 98.32% has been achieved for the training dataset, and 74.66%, on the test data. These results are comparable to results reported in earlier research but have been obtained without any feature extraction or post-processing.