A Method To Remove Haze In Images Based On Dark Channel Prior And Adaptive Fusion

The haze is a common weather phenomenon which benefits from dry aerosol(haze)and water droplets(fog),so “haze and fog” often appear together.Visibility under the hazy weather is very low,which seriously hinders the effectiveness of outdoor imaging equipment,resulting in blurred images and bad visual experience.More seriously,the impact of remote sensing equipment and UAVs on ground observations will hinder the subsequent detection and identification of targets;it will also affect the effects of urban traffic monitoring systems and cause frequent accidents.Therefore,the research to remove the haze and fog from outdoor images has important significance in both military and civilian fields.Methods to remove the haze and fog from images can roughly divided into two types: one is based on image enhancement and the other is based on physical model.The former is simple and direct,which improves visual effects by increasing contrast.The latter estimates the parameters in the model by some prior knowledge or constraints to get theoretical real images.The methods based on image enhancement have a fast processing speed but a large limitation,and excessive enhancement or color distortion tends to appear for unsuitable images.The methods based on physical model are more complicated but more universal,and it can achieve better recovery effects for different images.The method based on dark channel priors to remove the haze and fog of images stands out among many methods because of its effectiveness and excellent visual effects.But it has some disadvantages: First,for the sky area of smog images,the image processed by this method is easily over-enhanced and leads to halo generation because the dark channel prior is not applicable for sky region.Second,the solution of atmospheric light is not accurate enough.Last,the efficiency of this algorithm needs to be improved.Initially we used a simple hard threshold method to address the shortcomings of the algorithm based on dark channel prior.It has achieved certain results,such as the correcting the transmittance and atmospheric light,but the Information utilization is not high and the generality is not strong.In response to this problem,a method based on the dark channel prior and adaptive fusion was proposed.It can make the restored image clear and keep the sky area without distortion.The core idea of this paper is to combine the transmission rate calculated from the dark channel prior model and the transmittance calculated from the brightness model by a weight map to obtain the final transmission rate.Adaptive generation of weight maps depends on transmittance and sky segmentation.Among them,the transmission rate calculated from the dark channel prior model can effectively remove the haze in the foreground area of the images;the transmission rate calculated from brightness model can effectively respond to the sky area and ensure it natural.The solution of atmospheric light A depends on the detection of the sky region,and it is obviously more accurate to estimate the atmospheric light in the sky region than to solve the atmospheric light in the entire image.After the two parameters of the atmospheric scattering model are determined,the real image can be solved.The simulation results of the method mentioned in this paper are compared with various methods and it is found that there is a better defogging effect and the sky area is maintained well without any distortion such as halos.