Removal of fog from hazy images and their restoration

The prominent reason behind road accidents during the winter season is the presence of fog in the environment. Other important reasons for the degradation of visibility are haze, smog, clouds and rain. In the process of developing the automation of a vehicle on the road, visibility and contrast are the most affected parameters of the captured image or video. Road accidents can be prevented if images taken in foggy conditions are processed to improve their quality and legibility. There are different methods available to improve the quality of foggy images, like color attenuation prior method, dark channel prior method, and fog removal using region detection network.

The atmospheric particles, such as water droplets, which cause the absorption and scattering of light, further produce attenuation and air-light. The present research work is based on the Dark Channel Prior (DCP) method. The DCP method needs to find the transmission map, which gives the strength of the fog in the image. Major parts of this algorithm are the estimation of the dark channel, finding the transmission map, refining the transmission map, and reconstructing the image without haze. The proposed algorithm has also been implemented using a Raspberry pi. This research work focuses on the improvement of the reconstructed de-hazed image using various filters. The results are compared based on Contrast Gain (CG) and Color Index (CI) parameters. Many times, this application needs the object detection phase, which uses various methods; however, the scope of this paper is limited to the reconstruction of the image after the removal of fog.