Night Image Defogging Based On Multiple Light Source Models And Dark Channel Priors

Due to the influence of weather such as fog and haze,the optical imaging of the outdoor scene changes,resulting in serious deterioration of the imaging quality.In particular,photographs taken at night with poor light conditions exhibit low brightness,poor contrast,and color distortion.Although there are currently many excellent algorithms for image de-fogging that can restore images of outdoor natural scenes,most of the light from night-foggy images comes from artificial light.The classic image defogging algorithm is no longer applicable,so it is worthwhile for us to study how to construct a suitable night fog image imaging model.At present,researchers have proposed a model method for night image defogging,but the processing effect has problems such as color cast,loss of image details,and obvious halo.In view of these problems,this thesis makes improvements in the estimation of atmospheric light at night and multi-light source models to improve the above defects.Various models of night image defogging was fully studied and analysed in this thesis.And combied with the deficiencies in various methods to propose an improved night image dehazing algorithm based on multiple light source models.In this process,we first separate the original image from the texture layer and the texture layer;then we perform the dehazing operation on the structural layer.In the process of defogging,the image of the structural layer is also separated from the ambient light and only retained Part of the fog image;Finally,the part of the image through the dark channel prior algorithm to defogging,and then the image after defogging and texture layer fusion,so that the final restoration of the image details more clearly,significantly reduced halo phenomenon.Second,since the dark channel prior algorithm is used,atmospheric light values need to be estimated.The atmospheric light for the dark channel is estimated to have a non-uniform illumination in the night scene and a strong illumination phenomenon in the night image.We have improved the atmospheric light estimation of the night image.The estimation of atmospheric light value does not use the method of selecting the highest luminance value in the image to determine the atmospheric light value.Instead,it uses the method of optimizing the ambient light using the weighted regularization model based on the median filter to estimate atmospheric light.value.This improved method is used in the multi-light source model we proposed above.The experimental results show that the nighttime atmospheric light value estimated by this algorithm significantly improves the color shift phenomenon,laying the foundation for the following defogging process.