Unmanned Aerial Vehicle remote sensing image dehazing via global parameters

Abstract

Unmanned Aerial Vehicle (UAV) aerial mapping technology is becoming more and more widely used for its specific advantages. But in some areas, where there is fog or haze, the visibility of images degrades severely. To solve this problem, we propose a formation model for UAV remote sensing hazed images by converting the dehazing problem into the computation of two global parameters, including airlight vector and transmission. Specifically, we first get the airlight vector by a statistical law of variation in local image brightness, but only use its precise magnitude; secondly, we estimate the accurate airlight orientation based on the geometric property of pixels (projected in RGB space) of the patches in which the transmission and surface reflectance are approximately constant. Thirdly, we calculate a mean transmission from the top 20% darkest pixels in the dark channel and use it as the global transmission. The dehazing experiments based on synthetic images show that our algorithm can get a more accurate airlight vector and global transmission. The experiments based on actual UAV images show that our algorithm has a good histogram shape preservation, which can effectively enhance the clarity of the hazed images. For UAV hazed remote sensing images of mapping, our algorithm can effectively improve the visual effect and enhance visibility.