CNN-based data augmentation for handwritten gurumukhi text recognition

Models depicting deep learning have shown sustainable growth in recognizing handwritten words written in various languages, but the major challenges is faced in the field of image recognition and the collection of the dataset. To avoid such issues, the data augmentation technique is used for enhancing data to implement the various deep learning models. Typical computer models and DAR models through random elastic deformations and geometric transformations (such as shift and rotation) for initial instances of training. Through this research, an investigation on generative adversarial networks (GAN) has been carried and a method for generating fresh fake examples has been done. In the event of a training sample with a limited percentage of labeled examples that are represented in a high-dimensional space, the classifier could not translate well. Subsequently, developing a data augmentation technique for producing synthetic images, we seek to enrich image or signal databases to increase output performance for the defined classifier. This scientific research will help the researcher to understand how the data augmentation technique works and how it improves the performance of the deep learning models by enhancing the limited dataset. Our approach achieves high accuracy rates on the different feature extraction parameters by using a convolutional neural network on the enhanced data generated by implementing the data augmentation approach.