Image Haze Removal Based On Statistical Characteristic And Transmission Fusion

Images collected in the weather of fog,haze,weak light and other conditions will suffer from some problems such as low contrast,color distortion and vague details,which seriously affect the effectiveness of outdoor vision system.At present,a lot of progress has been made in the field of image dehazing in the daytime,while the study on the haze removal of images at night is less.Based on the research of the existing image dehazing algorithms,a new method of daytime haze removal based on deep learning and transmissivity fusion is proposed,besides a new method of nighttime haze removal based on statistical characteristics and brightness estimation is also proposed.The specific work is as follows:According to the imaging model for haze images,it is well known that obtaining accurate prior information is the key to improve the image defogging quality.At present,Various prior models can not fully and accurately characterize the fog related characteristics,resulting in limited applicability.In order to solve this problem,first,this paper studies the methods of image defogging based on MSCNN and DehazeNet,then the image defogging algorithm is proposed based on the fusion of deep learning and transmission which will fuse the transmissions obtained by deep neural network learning and the algorithm based on the dark channel prior,at the same time,our proposed method combined the direction of the atmospheric optical estimation,at last,the experimental results show that the proposed algorithm has many advantages with uniform in brightness,undistorted,and is ideal for fog removal in the sky area.Further,the paper studies the nighttime fog removal methods.Due to the existence of artificial light,nighttime fog image has the characteristics of non-uniform illumination,color cast and low intensity.Considering the character of the nighttime hazy image,a new imaging model for hazy image based on color cast factor is established,then the nighttime image dehazing algorithm is proposed based on the statistical properties of nighttime hazy images and the estimation of image intensity.Experimental results show that the proposed algorithm can remove the haze effectively,and at the same time the brightness,contrast and the image details are improved remarkably,while the computational complexity is very low.