Improved Image Super-resolution Using Enhanced Generative Adversarial Network a Comparative Study

Abstract

Super-resolution using generative adversarial networks is an approach for improving the quality of imaging system. With the advances in deep learning, convolutional neural networks-based models are becoming a favorite choice of researchers in image processing and analysis as it generates more accurate results compared to conventional methods. Recent works on image super-resolution have mainly focused on minimizing the mean squared reconstruction error and able to get high signal-to-noise ratios. But, they often lack high-frequency details and are not as accurate at producing high-resolution images as expected. With the aim of generating perceptually better images, this paper implements the enhanced generative adversarial model and compares with super-resolution generative adversarial model. The qualitative measures such as peak signal-to-noise ratio and structural similarity indices were used to assess the quality of the super-resolved images. The results obtained prove that, enhanced GAN model is able to recover more texture details when compared to super-resolution GAN models.