Fast Image Defogging Algorithm Design Based on Dark Channel Prior

Abstract

Foggy weather brings unfavorable influences to transportation, visual sensing, security and so on. Although some traditional algorithms have excellent processing effects on computers, it is difficult to implement on platforms with high real-time requirements such as self-driving and aerial photography. In this paper, one simple but efficient algorithm has been proposed to meet the requirement of real-time image defogging. Instead of dividing by squares, continuous plane detection strategy has been adopted so that one foggy image can be divided into many continuous parts marked by one independent label, avoiding the blocking effect of image processing. Based on Dark Channel Prior (DCP), estimate of atmospheric light and transmission function seems no longer troublesome. Meanwhile, downsampling has been used suitably to decrease the computational complexity of intermediate processes while avoiding loss of precision. Finally, we achieved much faster single image defogging speed as well as better PSNR and SSIM.