Pervasive multifaceted process based generative adversarial network for image quality enhancement

Abstract

The practice of Generative Adversarial Networks (GAN) has extended a lot of consideration in recent times. In most of the GAN methods are problem-specific related that are personalized to report numerous trials of individual application rather than performing other image improvement tasks. Furthermore, the basic GAN generators influence their boundaries in numerous image restoration and development use cases. Therefore, in this paper, we propose a generic GAN referred to as Pervasive Multifaceted Process based Generative Adversarial Network (PMPGAN). In this generator, we introduced multiple Convolutional Neural Networks (CNN) followed by Multi-dimensional Pyramid Pooling Module (MPPM) and Attention Module (AM), of which the input for the AM is given as low-level features and it produces output with enhanced feature map. Meanwhile, we enhanced the improvement outcome of the generated image with discriminator loss function. Finally, we tested the efficiency of the proposed system through extensive experiments on five challenging applications for image enhancement, image restoration, and infrared image translation to determine the dominance and efficiency in eliminating image degradation and producing visually interesting fake images. Our PMPGAN quantitatively outperforms several latest models. The results show that the PMPGAN model is superior to the existing models.