Multi-Scale Deep Residual Learning-based Single Image Haze Removal

Abstract

Images captured from outside visible devices are generally degraded by way of turbid media, including haze, fog, rain, and snow. Haze adversely degrades the quality of an image. Because of the weather conditions, haze is the most typical one in outdoor scenes. Due to the ill-posed nature of single images, dehazing them might be difficult. In this, a deep                  learning-based architecture that is denoted by MSRL Dehaze Net for single image haze removal is counting on multi-scale residual learning (MSRL). Haze removal in base can be achieved by multi-scale residual learning and simplified U-Net learning for mapping between hazy and haze-free components. The resulted dehazed image is obtained by the haze-removed base and also enhanced detail image components. This dehazed result passes through the clahe method for enhancement and obtained better final dehazed results. Experimental results have demonstrated and compared with other approaches such as Dark Channel Prior and Dehazing using GAN.