Abstract

Image synthesis is the process of generating new images from the ground up, frequently utilizing preexisting data or models. By definition, super-resolution methods produce supplementary image content and features that are not present in the original input in order to reconstruct a high-resolution image from a low-resolution source. Surpassing the performance attained with high-resolution images is a challenge when training or analyzing models with low-resolution images. It is not always easy to obtain a higher-resolution image. Recognizing and identifying objects in low-resolution images is a challenging task. As a result, it is imperative to develop a method that simultaneously enhances the resolution of the low-resolution image and enhances its quality. Generative Adversarial Networks (GANs) and other generative models are increasingly acknowledged for their capacity to accurately replicate high-resolution counterparts. This article provides a performance comparison of DCGAN and Wasserstein GAN for image synthesis via image super resolution. The experimental endeavor employs the Set5 image data set. Wasserstein GAN outperforms DCGAN in terms of PSNR, SSIM, and VIF parameters.