Enhanced Video Dehazing using Deep Learning and Haze Density Guided Region Attention.

Haze significantly degrades the visual quality of videos, hindering the performance of various vision-based applications. While deep learning techniques have shown promise in video dehazing, many existing methods apply a uniform dehazing effect across the entire video frame, often leading to over-enhancement in clear regions and insufficient dehazing in hazy regions. This paper presents an adaptive video dehazing method that addresses this limitation by incorporating a Haze Density Estimation Module. This module analyzes the input video frames to estimate the spatial distribution of haze. The estimated haze density maps are then used by a modified Region Attention Module to adaptively focus the dehazing process, applying stronger dehazing in denser regions and preserving details in clearer regions. Experimental results demonstrate that the proposed method effectively removes haze while preserving image details, outperforming existing global dehazing approaches both qualitatively and quantitatively. This adaptive approach offers significant potential for enhancing video clarity in applications where haze is non-uniformly distributed.