Effective edge-aware weighting filter-based structural patch decomposition multi-exposure image fusion for single image dehazing

• Published in: Multidimensional systems and signal processing Vol. 34; no. 2; pp. 543 - 574
• Main Authors: Yadav, Sumit Kr, Sarawadekar, Kishor
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.06.2023; Springer Nature B.V
• Subjects: Artificial Intelligence; Circuits and Systems; Computer vision; Decomposition; Edge detection; Electrical Engineering; Engineering; Image filters; Image processing; Remote sensing; Signal strength; Signal,Image and Speech Processing; Weighting; Transmission map; Structural patch decomposition (SPD); Guided image filter; Gradient domain; Multi-exposure image fusion; Gamma correction
• ISSN: 0923-6082; 1573-0824
• DOI: 10.1007/s11045-023-00873-z

Image dehazing is a severe and challenging problem due to its ill-posed behavior and is highly desired in various vision based applications such as computer vision, image processing, computational photography, remote sensing, outdoor driving assistance, and video surveillance, etc. Therefore, we proposed a novel effective edge-aware weighting filter-based structural patch decomposition multi-exposure image fusion method for single image dehazing. It removes haze firmly and preserves edge information precisely in the flat and sharp regions. The proposed method comprises four steps. First, a set of four gamma coefficients ( γ = 2 , 4 , 6 , 8 ) is applied to the input hazy image and obtained the corresponding underexposed outcomes, respectively. Then, the structural patch decomposition method decomposes each underexposed image into three independent elements: signal strength, signal structure and signal mean intensity. Next, a novel, effective edge-aware weighting-based guided image filter is used to refine each decomposed image patch. Finally, these refined images are fused to achieve a compelling haze-free image. The proposed method removes halo artifacts, over smoothing and color cast strongly, and preserves edge information precisely in both flat and sharp regions. The theoretical analysis and tested outcomes prove that the proposed haze removal method can produce faster and more effective outcomes than the existing haze removal methods.