Research On Thin Cloud Removal For Single Aerial Images

In recent years,with the continuous progress and popularization of remote sensing equipment,remote sensing technology gradually becomes mature,and is widely used in space exploration,global environmental monitoring,crop yield estimation,weather forecasting and military reconnaissance and other fields.Although remote sensing images carry a large number of high-resolution information,it is easy to be affected by bad weather factors such as haze and cloud,causing a part of the information on the remote sensing image to be obscured and bringing serious interference to the efficient use of remote sensing images.If the clouds on remote sensing images can be eliminated,the integrity of remote sensing images could be restored and their transmission utilization and utilization value should be improved.The cloud removal process eliminates the cloud interference and restores the occluded ground information on an cloudy interfered remote sensing image,and it essentially belongs to image restoration problem.Due to the different periods and different locations,the texture and concentration vary seriously.Its shape and characteristics are subject to different climate and seasonal effects,and clouds show a great difference in appearance.Most of the existing cloud removal technologies need other band information captured by remote sensing equipment equipped with special sensors.The input is cumbersome and the processing is extremely complex and difficult to apply to general RGB images.If only a single-scene RGB image is used as input,it would be very difficult to completely remove the cloud information on the entire image because the image does not provide any information for the objects covered by thick cloud,so the task of this paper is to removal as much thin clouds as possible on remote images.In this paper,we propose a new model-based cloud removal method for a serious of problems appearing in the process of cloud removal on single RGB remote sensing images.The key to our method is a spatially varying atmospheric light map which can adjust its values according to the thickness of cloud.In this paper,we first construct a significance map which highlights the concentration of the thin clouds.Based on the significance map,we generate the rough atmospheric light map with the brightest pixel of dark channel.We then optimize the rough atmospheric light map by a L1-norm based regularization method.We further calculate the transmission map directly from the input image and its atmospheric light map.For refining processing,we use the RGB color map as guidance image for the guided filter.Finally,we restore the image with a mask result of a cloud detection method which speeds up our algorithm dramatically and keep the original global tone.The main contributions of this paper are as follows:1.The input data of our algoritlum is a simple RGB color aerial image,and we can automatically remove most cloud with no additional band information and user interaction.2.Our algorithm flexibly utilizes cloud detection technology.Processing only the cloudy pixels using cloud detection result as a mask significantly reduces run time of the operation.3.We develop existing method using a spatial varying atmospheric light map a,and first apply it to the thin cloud removal field.In this paper,related works of thin cloud image removal technology are studied deeply,and a lot of experimental comparison results are provided.The results of multiple sets of images prove the feasibility and effectiveness of the thin cloud elimination algorithm.