Image Generation Using GAN and Its Classification Using SVM and CNN

In the computer vision world, generative adversarial networks have acquired huge recognition because of their data generation potential without modeling the function of probability density. Generative adversarial networks (GANs) have capability to generate new samples similar to data they were trained on. A smart means of integrating samples without label into preparation and applying higher-order accuracy is adversarial loss generated by discriminator. In this paper, GAN is implemented on MNIST dataset and recognizes latent representation of the feature for digit generation. The model also consists of support vector machine (SVM) which is initially trained on the same MNIST dataset that is being used by GAN for generating new data similar to that of MNIST. The data generated by GAN are then passed through the pre-trained SVM classifier for predicting its equivalent label. The support vector machine model which was used with the MNIST dataset obtains the accuracy of 96.67%. Moreover, the proposed model with convolution neural network (CNN) on the sameMNIST dataset produces an accuracy of 99.22%.