Enhanced Deep Learning-Based Image Dehazing Technique and Comparative Analysis

Typically, atmospheric variables including dust, smoke, fog, haze, snow, and rain reduce the effectiveness of high-resolution cameras. Degradation can affect the color, contrast, and visual quality of an image. Degraded or hazed images are unsuitable for real-time image processing applications because they lack information specifics. For vision applications, the dehazing step with respect to the images is compulsory to enhance the quality. The primary goal of dehazing is to maintain visual quality while retaining information and scene clarity. An architectural framework for deep learning-based image restoration and single-image dehazing is presented in this work. We propose to reframe the challenge as restoring the image's base component, as opposed to learning a direct mapping between hazy images and their haze-free counterparts, as is the case with most existing approaches. Haze can be removed from a hazy image by first breaking it up into its basic and complex components. Then, by training a CNN to connect the hazy and haze-free fundamental components, additional enhancement of the complicated component can be achieved. The components of the initial image, the haze removed, and the enhanced detail image are combined to generate the final dehazed image.