Research On UAV Remote Sensing Image Dehazing Algorithm

UAV aerial photography, mapping and other technologies, are becoming more and more widely used for their specific advantages. But in some areas, where there is fog or haze due to weather conditions, atmospheric pollution,or other factors, the images collected can become severely degraded,which greatlyaffects the follow-up works such as image analysis and understanding. So it has important practical significance to dehaze UVA remote sensing images got in fog days. For this problem,we have done some works as following.By comparing ordinary images of scenes and UAV remote sensing images, we summarize the characteristics of UAV remote sensing image, thenproposea UAV remote sensing image dehazingmodel, and convert the dehazing problem into the computation of the air-light vector and global transmission.Firstly,we get the air-light by a weak statistical law of localimage brightness, Because the orientation of the air-lightis not very accurate,we only use themagnitude;Secondly,we search a sufficient number of small image patches in which the transmission and surface reflectance are approximately constant,then we estimate a accurate air-light orientition by the geometric property of the pixles in patches when they are projected into RGB space.Thirdly,calculatethe mean transmission of the bottom 20% brightness pixels in dark channel, and use it as the global transmission.To verify the feasibility and the actual dehazing effect of the proposed algorithm.On the one hand,we give a method of constructing synthetic images,and then do a test with proposed algorithm and some other popular algorithmsbased on these synthetic images with known atmospheric light vector and transmittance.The results show that our algorithm can get more accurate air-light vector and global transmission.On the other hand,using the algorithms above,we remove haze from two actual UVA remote sensing images.The results show thatthe image dehazed by our algorithm has better matching degree and histogram similarity than the ones processed by the other algorithms,and our algorithm can effectively enhance the clarity of the hazed images.