Single Image Defogging using Deep Learning Techniques: Past, Present and Future

Image dehazing play a vital role in several applications related to computer vision. The prime motive of this paper is to provide the overview of the existing deep learning algorithms associated with image defogging. In beginning, the main issues preset in the existing single image technique based on physical models are discussed. Thereafter, the basic concept of atmospheric scattering model and deep learning are discussed. The existing deep learning approaches based on single image defogging are decomposed into 3 broad categories namely Convolution neural network, Recurrent neural network, and Generative adversarial network with their pro and cons are discussed. The synthesised and real datasets used in defogging techniques are discussed with their applications. It also describes the various challenges and issues in the existing image dehazing techniques.

Automated summary

This paper provides an overview of existing deep learning algorithms for image dehazing, discussing issues with current techniques and the basic concepts involved, as well as analyzing different deep learning approaches and their applications. It also highlights challenges and problems in existing image dehazing techniques.