Fast Image Dehazing Based On Feature Space Sampling And Recursive Bilateral Filtering

When combined the water vapor with dust in the air,it will form more and more tiny water droplets,which suspended in the air lead to the existence of fog.But the emergence of the fog would lead to a blurry image.In foggy weather conditions,the image is inevitably scattering by aerosols in the atmosphere,so that the contrast is not high,and color is distorted,and the detail and edge is not complete.So in such a case the image obtained will brings to the inconvenience of the users of the image,which have a serious influence on our daily life.And based on the theoretical knowledge of image processing,the Image dehazing algorithm can recover the real scene from the foggy image,and get more useful information,and improve the visibility of the image,and facilitate the subsequent processing of the image.Image dehazing is an research branch of image processing,at the same time it is a specific application of the image enhancement technology,and it is also a slightly different from image enhancement technology,that it has the characteristics of image enhancement technology,but also it has its distinctive place,which it can improve the quality foggy image.Although many dehazing algorithms have been proposed,but most of them have some limitations,so there is still a lot of research space in this field.In this paper,according to the different fog in the image,we used two different dehazing algorithm to get a clear image without fog.In the mist image,the fog distribution was more homogeneous in the image,and the edges of the objects in the scene were relatively clear.In this article,we applied the method of the medium transmission function to remove the fog.At first,we estimated the initial value of the medium transmission function.Then recursive bilateral filtering was applied to refine the initial value of the medium transmission function.Finally,based on gradient domain recursive bilateral filtering,we adjust the dehazing result.In order to accelerate the dehazing algorithm,based on Gaussian KD tree,we applied feature space sampling to obtain the high dimensional feature space of the foggy image.Compared with existing methods,this method can effectively to dehazing the mist image,and edges and details were keeping well.In the heavy fog image,we applied the atmospheric veil method to remove fog.We first obtained the initial atmospheric veil.Then,we performed the recursive bilateral filtering to refine the initial atmospheric veil for avoiding inducing halo artifacts,in order to get clear images of high quality.To accelerate the image dehazing,the affinity propagation algorithm was applied to obtain the sampling of feature space in the foggy image.Compared with the exiting dehazing approaches,the method could get better dehazing results for the haze images with many edge features and complex scenes.And based on several neighboring frames information around current frame of the video-image,the globe atmosphere light of current frame was estimated,and the atmospheric veil or the medium transmission function was filtered across each frame of the video-image with the recursive bilateral filtering operator.Thereby these method were extended to video-image,which obtained the video-image dehazing result with temporal coherence.