A Study On Single Image Dehazing Algorithm Based On Atmospheric Scattering Model

Now,computer vision system has been widely used in navigation,tracking,surveillance and other security fields,and the image quality acquired via it has a significant impact on the task,which is whether the targets of interest can be accurately located or identified.However,in foggy weather conditions,the images captured by the visual system are usually blurry and low in color saturation,which will ultimately directly limit the performance of the overall system,and which is also a hidden trouble for people's life and safety.Therefore,in order to obtain clear images,it is necessary to study the effective image dehazing algorithm,which is also a difficult task.About the image degradation problem above,based on previous studies on atmospheric scattering model and relevant theoretical analysis,this paper is mainly to conduct research from the following two aspects:(1)An efficient dehazing method is proposed for a single input image by combining the dark channel prior information and a low-light image enhancement model.First,the rough scene transmission estimation is derived from the dark channel prior information,but rough estimates of transmission in the region of the sky are often subject to considerable error,in order to achieve fine estimations of the sky area,this paper continues to modify the sky region with the transmission obtained from the boundary constraint condition estimation,and then obtains a relatively accurate initial transmission.Then,this paper employs a low-light image enhancement model to refine the initial transmission to obtain the final transmission.At the same time,the reasonable atmospheric light is estimated through dark channel prior in this paper.Finally,with the estimation of atmospheric light and transmission,the clear image is reconstructed based on the atmospheric scattering model.(2)An improved dehazing algorithm based on the color attenuation prior is proposed for a single input image.First,when estimating the scene depth information by using the color attenuation prior,this paper employs the bilateral filtering to further smooth the block effect caused by the minimum filtering operation,so as to obtain the relatively uniform initial transmission.Furthermore,the context regularization based on weighted L1 norm is used to construct the objective function to refine the initial transmission.Then,our work firstly picks the top 0.1 percent brightest pixels in the estimated scene depth map,and the one with the highest intensity is selected from the corresponding pixel of the original image as the estimated value of atmospheric light.Finally,the scene irradiance can be recovered according to the physical model and estimated model parameters.The above two algorithms are compared subjectively and objectively with the existing algorithms through experiments.The experiments show that the proposed algorithm is effective.