PROJECT TITLE :

Semi-Supervised Image Dehazing

ABSTRACT:

The dehazing of a single image can now be accomplished using a Semi-Supervised Learning system that we've developed. A deep Convolutional Neural Network (CNN) with supervised and Unsupervised Learning branches is used in the approach. Deep Neural Networks are bound by the supervised loss functions, which include mean squared, perceptual, and adversarial losses, when working in the supervised section. It's in the unsupervised branch that we use image attributes like light channel sparsity and gradient priors in the network. " An end-to-end training approach is used to develop the suggested network using both synthetic and real-world data. Analysis of the proposed semi-supervised learning technique indicates that it is not confined to synthetic training datasets and can be easily generalised to real-world photos. On benchmark datasets as well as real-world photos, the proposed approach outperforms the current state-of-the-art single image dehazing algorithms.