The Study Of Haze Removal Algorithm Based On Single Image Containing Sky

In recent years,environmental pollution has led to heavy haze in the Northern China.The images obtained in this region are dusky and fuzzy,which creates problems for working on those images,for example,target detection,recognition and track in aerospace by the devices such as unmanned aerial vehicles(UAV),and taking the aesthetic pictures by mobile phones in daily life.Therefore,the study of image dehazing has its important practical significance.In order to deal with the haze and dim,algorithms have been developed based on the studies on close-range-shot processing.However,in long-range-shots,the far distant area(background)and sky area are difficult to restore.The irradiance in long-range shot is attenuated by much more turbid mediums in atmosphere.There is no standard to estimate its thickness,which causes the restoration of long-shot images to become a challenging problem.Taking into account these issues,in this paper,we propose two new single image haze removal method based on sky region.The main work of this paper includes:(1)We proposed a haze removal based on soft segmentation method to effectively restore long-shot images with sky.The transmission values estimated based on a luminance model and dark channel prior model are fused together based on a soft segmentation.The transmission estimated from the luminance model mainly contributes to the sky region,while that from the dark channel prior for the foreground region.The airlight also is adjusted to adapt to real light by sky region detection.A user study and objective assessment comparison with a variety of methods on long-shot haze images demonstrate that our method retains visual truth and removes the haze effectively.(2)Most existing haze removal methods cannot restore a clear sunny weather scene,The issue of image haze removal has attracted wide attention in recent years.However,most existing haze removal methods cannot restore a clear sunny weather scene,since the color and texture information of the object in the original haze image is insufficient.To remedy this,we discard the atmospheric physics models and use generative adversarial network for end-to-end dehazing.We use outdoor image datasets to train our model,which includes a set of real-world unpaired image dataset and a set of paired image dataset to ensure that the generated images are close to the real scene.Based on the cycle structure,our model adds four different kinds of loss function to constraint the effect include adversarial loss,cycle consistency,photorealism regularization and L1 paired data loss.These four constraints can improve the overall quality of such degraded images for better visual appeal and ensure enhanced restrict and reconstruct image to keep from distortion.The dehaze generator can produce an image with clear and bright blue sky.The experiments show that our method outperforms the state-of-the-art methods.