The achievement of any Deep Neural Networks will be depends of the training data set of images. But for certain problem the datasets which are collect from different resources are not sufficient to do classification of images with high accuracy. For such type of image classification works, it is essential to use a variety of data enlargement methods for the inadequate training image samples to enough training samples. The present paper explores the different data augmentation techniques impact on image classification works with Deep Neural Network on Handwritten Hindi digit dataset from Devanagari script. Dissimilar data augmentation techniques used in this paper contains rotate, Zoom, horizontal shift and Vertical shift. This sample data augmentation method considerably enhanced classification accuracy for every tested datasets. This method is most valuable for these works with a incomplete amount of training data, such as medical imaging tasks.
Volume 11 | Issue 12
Pages: 90-96
DOI: 10.5373/JARDCS/V11I12/20193216