Refining Transmission Map and Air Light for Efficient Single Image Dehazing

Abstract

Many dehazing methods have been presented by many researchers either on single hazy images or multiple images like videos. Single image dehazing is a very challenging problem, since there is no ground truth image and/or depth map. Beyond many traditional dehazing methods, with the increasing volume of hazy image datasets, recent literature about dehazing mostly focuses on deep learning methods. Although, deep learning models perform better than traditional methods on the validation data of the specific dataset which they are trained with, the generalization performance of deep learning models is generally poor. Another important bottleneck for deep models is the training cost in terms of time and hardware requirements. In this study, Atmospheric Light Scattering model is improved by taking into account the amount of haze and refining the transmission and air light used in atmospheric light scattering model. According to VIS quality metric results, proposed method is the second best method on Dense-Haze and CHIC-Color datasets by achieving 0.87 and 0.86 scores, respectively, and the best method on our own IMFD dataset by handling 0.91. Results of this study shows that proposed method is superior and/or competitive to the state of the art deep learning models in terms of the visual quality of the dehazed image.

Keywords

• Atmospheric effects
• foggy image
• image reconstruction
• visibility distance