1. SANGITA ROY - Associate Professor at ECE Department, Narula Institute of Technology.
2. SANDHYA PATTANAYAK - Associated with Narula Institute of Technology, Dept. of Electronics & Communication Engineering.
3. PRANAB HAZRA - Associated with Narula Institute of Technology, Dept. of Electronics & Communication Engineering.
4. KAUSHIK SARKAR - Assistant Professor, in the Department of Electronics and Communication Engineering of Narula Institute
of Technology.
5. SURAJITBARI - Associated with Narula Institute of Technology, Dept. of Electronics & Communication Engineering.
6. ABHIJIT GHOSH - Assistant Professor in Electronics and Communication Engineering Department, Narula Institute of
Technology.
7. ANILESH DEY - Associate Professor of Electronics and Communication Engineering at Narula Institute of Technology.
A novel method has been proposed for a single image dehazing technique to dehaze both daytime and night-time hazy scenes. Inverting the popular Koschmieder optical image formation model (KOIF) [10] with dark channel [18], the airlight on image patches is light, but large patches for accurate airlight estimation by increasing the possibility assessed and not on the entire image. Local airlight estimation is incorporated for night-time conditions with the nonuniform lighting from multiple localized artificial sources [32]. Patch Size selection is significant, small patches for fine spatial adaptation to atmospheric of capturing pixels with airlight appearance (due to severe haze). To alleviate the said problem, airlight is estimated as the brightest pixel from medium order statistic filter (MOSF) refined transmission map. The depth map is improved with the Minimum Order Statistics filter (MOSF) [37], which in turn improves the transmission map. Finally, a clear image is derived by inverting the KOIF model. The radiance is improved with a low light image enhancement technique [31]. Extensive experimental results established the effectiveness of the proposed approach as compared with recent techniques, both in terms of computational efficiency and the quality of the outputs.A novel parallel atmospheric light and depth map estimation concept has been implemented for faster operation.
Airlight, haze, dehazing, MOSF, KOIF, Image Formation Optical Model, MOSF, PSNR, SSIM, NIQE, BRISQUE.