Research On Single Image Haze Removal Using Dark Channel Prior

In recent years,haze weather occurs in most parts of China frequently.It not only affects people's lives seriously,but also there are some disadvantages such as high gray value,low contrast,low signal to noise ratio and so on under such weather conditions,which influence the subsequent image processing.It is very important to defog the fog image through related methods.The dark path prior theory provides a new idea for image defogging.The image defogging method based on dark channel priors has better defogging effect,but there are some deficiencies such as high time complexity of the algorithm,the color of the restored image sky area is distorted and so on.Therefore,the improvement of single image defogging algorithm based on dark channel priors is studied.The main research content of the thesis is as follows.Firstly,the characteristics of atmospheric scattering theory and haze weather are analyzed,and the imaging principle and process of fog images is studied.Then introduced the physical model of atmospheric scattering.Through the further study of the two-part model that composes the physical model of atmospheric scattering,the key to the image defogging based on the atmospheric physics model is obtained.Then atmospheric light and transmittance are calculated by combining the dark channel theory and the atmospheric scattering physical model,and the rough estimate of the transmittance is refined by the soft mating method.Finally,complete the fog image recovery according to the physical model of atmospheric scattering.The results show that the refinement of the transmission rate by the soft matting method leads to an excessively high time complexity of the algorithm,the sky region of the restored image is distorted,and the edges of the restored image appear blurred.Therefore,this thesis proposes an improved method for the lack of traditional dark channel prior defogging algorithms.The main idea is to use two dark channels of different sizes to obtain the rough estimate of the transmittance,use guide filtering to optimize it.And determine whether the fog image contains a sky area,propose a coefficient compensation for the transmittance of the sky region.The guide filter is used again to optimize the transmittance and get the final transmittance.At the same time,according to the quadtree multi-level search method,the accurate estimate of the atmospheric light is obtained with the dark channel image.Finally,the final restoration is achieved through the atmospheric physics model.The defogging experiments were compared respectively for the foggy pictures containing the sky region and not including the sky region.The results show that the algorithm has the following advantages: 1.The running time of the algorithm is reduced by 90% compared with the soft matting method;2.Eliminates the problem of color distortion in the sky region when the dark channel is used to defogging a fog-containing image in the sky region;3.The situation that the edge of the image caused by the leading filter is blurred is improved,and the fog between the detail edges in the restored image is removed.