Dehazing Algorithm Research For Haze Image Based On Statistics

Image processing is designed to highlight some of the details of the image in order to facilitate visual observation of the human eye and computer follow-up analysis operations.In the haze environment conditions,due to the role of atmospheric particles,outdoor image visibility is limited and the image contrast has declined.In order to solve the problem of reduced image quality of foggy weather,most of the current processing schemes are based on image enhancement and image restoration to eliminate the fog algorithm,image enhancement algorithm by improving the contrast of the image to achieve clarification effect;Image restoration based on the fog image imaging model,modeling the atmospheric scattering mechanism,to recover the foggy clear image.This paper mainly analyzes the physical model and the degradation mechanism for image defogging,and explores the key techniques and implementation and gives the dark channel prior statistical evidence and non-local defog improved algorithm.The main works of this paper are shown as follows:(1)The dark channel prior is a kind of statistics of the haze-free outdoor image,which is the most of the outdoor foggy images of the non-sky local area there are at least one color channel with very low intensity pixels.This article from the statistical point of view,assuming that the three channels are independent of each other,the scene points are independent of the other pixels in their domain,and treated the RGB values as the statistical variables which are homogeneous after color foggy images.Assuming that these three variables are subject to the beta distribution,the density function and the distribution function of the two variables after the minimum filter(First to the RGB three-channel small,and then take a small neighborhood)are given to verify the validity of the dark channel a prior.(2)According to the distinct prior information,the existing image haze removal methods can be divided into local and non-local.Berman et al.construct a geometric representation(haze-line)of each color class of haze image which is based on the non-local clustering characteristics of clear image in the RGB space.The largest radial coordinate(LRC)of haze-line is the key to estimate the initial transmission.From statistical point of view,this paper proposes an unbiased estimate of LRC.The experiments show that the proposed method can gain at least comparable results with original method.