Nighttime Image-Dehazing: A Review and Quantitative Benchmarking
- PDF / 4,828,695 Bytes
- 33 Pages / 595.276 x 790.866 pts Page_size
- 63 Downloads / 172 Views
ORIGINAL PAPER
Nighttime Image‑Dehazing: A Review and Quantitative Benchmarking Sriparna Banerjee1 · Sheli Sinha Chaudhuri1 Received: 28 January 2020 / Accepted: 21 August 2020 © CIMNE, Barcelona, Spain 2020
Abstract Visibility enhancement of images captured during hazy weather conditions is highly essential for various important applications like intelligent vehicles, surveillance, remote sensing, etc. In recent years, researchers proposed numerous imagedehazing methods mostly focusing on daytime images’ characteristics. In this work, we have highlighted the dissimilarities among the characteristics of daytime and nighttime hazy images and explained that well-known daytime image-dehazing priors cannot dehaze nighttime hazy images effectively. Following this discussion, we have provided a comprehensive review of existing nighttime image-dehazing methods after grouping them according to different nighttime hazy image models based on which they were designed as their methodologies vastly vary with those models. Thereafter, we have performed comparative qualitative and quantitative analyses of outputs obtained by applying these methods on images belonging to novel N-HAZE database. N-HAZE comprises of both indoor and outdoor real-world nighttime hazy images captured in the presence of haze created by artificial haze machines and corresponding Ground Truth images. Finally, we have concluded our work by stating the existing challenges and future scope of work in this field after analyzing the strengths and limitations of each method. Our main aim behind conducting this survey is to draw the attention of more researchers towards this less explored yet significant research topic and encourage them to design new methods which can solve the existing challenges. To the best of our knowledge, we are the first ones to review the nighttime image-dehazing methods and to design N-HAZE, which is the first database designed for benchmarking these methods. Keywords Nighttime image-dehazing survey · Dark Channel Prior · Spatially varying illumination characteristics · Glow characteristics · Nighttime hazy image models · N-HAZE database Abbreviations NHTSA National Highway Traffic Safety Administration DCP Dark Channel Prior GT Ground Truth SSIM Structural Similarity Index MSE Mean Square Error PSNR Peak Signal-to-Noise Ratio FADE Fog Aware Density Evaluator GIF Guided Image Filter CCT Color Channel Transfer MRP Maximum Reflectance Prior DF Dilation Factor * Sriparna Banerjee [email protected]; [email protected] 1
Sheli Sinha Chaudhuri [email protected]; [email protected]
ETCE Department, Jadavpur University, Kolkata 700032, India
BCP Bright Channel Prior HDP Haze Density Prediction ReLU Rectified Linear Unit HDP-Net Haze Density Prediction Network
1 Introduction Visibility of images captured during hazy weather conditions get severely degraded due to scattering and attenuation of scene light by fog, mist, dust and other aerosol particles present in the atmosphere. Hence, proper rest
Data Loading...
