Performance analysis of classification of handwritten Devanagari digits using new augmentation technique of squared patterns and DCNN model

Handwritten script recognition has risen as a significant issue in a previous couple of years. Areas of computer vision and pattern recognition has attracted many researchers for a lot of real-time applications. Among these fields, Devanagari script recognition is a hot area for many Indian researchers due to its many applications like document reading, postal address sorting, etc. Lot of deep convolutional neural network (DCNN) models are being implemented to increase recognition accuracy. The objective of this study is to compare the performance of the proposed DCNN model by applying two image formats that are without processing (original) and with our proposed augmentation technique. The results show that the classification accuracy has increased by using the proposed technique of augmentation on a very limited training benchmark dataset of handwritten Devanagari digits proposed by Dongre and Mankar. The authors have achieved the highest classification accuracy of 98.49%.