Research On Image Dehazing Methods Based On Dark Channel Prior And Modified Model

In the hazy environment,the image quality of the scene is seriously degraded,resulting in the failure of outdoor vision systems such as traffic monitoring,automatic driving,and automatic inspection robots to obtain clear target information.In order to get a clear image,the research of dehazing method is particularly important.Aiming at the problems of halo artifacts,color distortion,low image brightness,color deviation and glow in fog at night,corresponding improvement methods are proposed.The main research contents are as follows:(1)After the traditional dark channel prior algorithm dehazing,halo artifacts will appear at the abrupt change of depth of field,and serious color distortion will occur in the sky region.To solve this problem,an image dehazing method with improved dark channel window and transmittance correction is proposed in this paper.The image is processed by the superpixel segmentation algorithm to obtain a local window with the same depth of field.Using this window to replace the square filtering window in the dark channel can effectively eliminate the halo artifacts.Adaptive tolerance is proposed to correct transmittance in bright region,which can effectively suppress distortion in sky region and make image more natural.The experimental results show that the image obtained by this algorithm has clear details,natural colors,and can handle multiple types of hazy images with better robustness.(2)The hazy environment at night is complex,and the traditional atmospheric scattering model does not conform to the characteristics of optical imaging at night,so it cannot be used for night haze removal.To solve this problem,a new nighttime haze imaging model is established based on the imaging characteristics of scene and atmospheric light.Add the surface illumination light component to the attenuation term,add glow imaging to the atmospheric light term,and introduce the near light source coefficient matrix to establish the relationship between glow and atmospheric light.Based on this model,the corresponding night image dehazing method is further proposed.First,the glow is calculated according to the low-frequency characteristics.Then,after removing the glow,the dark channel prior algorithm is used to remove the haze,and the Retinex algorithm is used to remove the illumination component to restore the true color of the image.Finally,the image definition is further improved by enhancing details.A large number of experiments show that the proposed method can effectively restore image color,improve image detail and brightness,and is superior to the contrast algorithm in various indicators.(3)In order to test the dehazing method and the practical application of simulated image dehazing in intelligent devices,an image dehazing system is designed using Matlab,which integrates the daytime and nighttime image dehazing algorithms proposed in this paper.The system has three dehazing modes: automatic dehazing,daytime dehazing and nighttime dehazing.Automatic dehazing can automatically judge the image type according to the user’s choice of image,so as to select the corresponding image dehazing algorithm.When the system is not accurate,users can choose the corresponding dehazing method.Through this system,users can evaluate the dehazing algorithm more conveniently.