Development of Enhanced Network Model for Image Dehazing Using Combined Transmission Map Estimation

The elimination of haze from a single image presents a significant challenge due to its fundamentally ill-posed characteristics. Numerous prior-based and learning-based approaches have been introduced in the literature to address this issue, yielding visually pleasing outcomes. Most existing methods, however, presume a constant atmospheric light model and typically adhere to a two-step process: first, employing prior-based techniques to estimate the Transmission Map (TM), subsequently the computation of the dehazed image using a closed-form solution is carried out. This research relaxes the assumption of continuous atmospheric light and introduces a revolutionary unified single-image dehazing network that concurrently estimates the TM and executes dehazing. This novel approach offers a comprehensive learning framework, wherein the intrinsic TM and dehazed outcome are concurrently learned from the loss function. Comprehensive studies conducted on simulated and actual datasets featuring challenging hazy images concluded that the suggested method significantly outperforms existing techniques.