Tackling Multiple Visual Artifacts: Blind Image Restoration Using Conditional Adversarial Networks

Abstract

Restoring images that are degraded by visual artifacts like noise, blurness and other environmental visual artifacts like shadow, snow, rain, and haze is a challenging task. From literature, it can be seen that there are many model-based as well as blind restoration methods that have been proposed to restore an image degraded by a single artifact. In most practical cases, the image is degraded by more than one artifact. Complexity arises while trying to estimate degradation function using conventional techniques where images are degraded by multiple visual artifacts. To the best of our knowledge, there has not been any generalized method proposed to tackle this problem. In this paper, we propose a methodology using conditional adversarial networks for blind image restoration of images that are degraded by multiple artifacts. To analyze the performance, ISTD dataset (meant originally for shadow removal) is used by augmenting it with different types of noises and blurness. The network has been trained on this data and has been analyzed how it behaves during the addition of each artifact. Various image quality metrics like Peak signal-to-noise ratio (PSNR), Mean squared error (MSE), Structural similarity index (SSIM), Blind/reference-less image spatial quality evaluator (BRISQUE) and Naturalness image quality evaluator (NIQE) have been evaluated to validate the performance of the proposed method.