Research On Single Image Defogging Algorithm Based On Atmospheric Scattering Model

Nowadays,haze and fog frequent weather conditions.Under these conditions,the outdoor scene images captured by the optical equipment are seriously degraded due to the scattering of atmospheric particles,showing dynamic range decrease,low contrast,color distortion,and sharpness dropping.Therefore,the identifiability of the information is greatly reduced,unable to meet the needs of areas such as road monitoring and military reconnaissance.The reduction has a tremendous impact on all aspects of production and life.As the key to improving image quality,image de-haze technology has very important significance and value.Therefore,this thesis proposes three image de-fogging algorithms based on atmospheric scattering model.The main work of this thesis includes the following three aspects:(1)The improved dark channel prior algorithm based on local adaptive template has been proposed.First,introduce adaptivetemplate to the traditional dark channel prior algorithm.In a square neighborhood centered at the current pixel,the connection lines of the central pixel to each point on the square border are the initial direction.Bresenham algorithm is used to extract the pixel points in each direction and the length is determined by the similarity function,thus a more accurate dark channel prior color template is achieved,and is used to calculate the transmittance;for transmission refinement,the HSV-based color image gradient is used to build the exponential weight to penalize the normalization factor in the Guided filter,effectively highlighting the edge and texture detail.Experimental results show that this algorithm has a good effect on avoiding the halo effect and fog residual.(2)The de-haze algorithm based on color attenuation a priori and haze measurement has been studied.In a haze image,the higher the concentration of fog pixels is,the higher the brightness and the lower the saturation.Thus,the scene depth of the foggy image is modeled,the model parameters are obtained by the supervised learning method,the scene depth information is restored,and the foggy image transmission is obtained.Then,a quantitative map of fog concentration is established according to the brightness and saturation of the image and the color image gradient map.Both of them are used as a two-dimensional feature of SOM neural network training,the most hazy area is obtained by clustering,which is used as a candidate area of atmospheric light value.Experimental results show that this algorithm can effectively restore the scene visibility and improve the color distortion of the sky and bright area.(3)The night-time image de-haze algorithm based on layer decomposition theory has been studied.Firstly,a glorification term is added to the direct attenuation term and atmospheric term of the existing atmospheric scattering model.Then,a haze image is obtained by separating and removing the glitter layer from the input image using a layer decomposition algorithm based on relative smoothness.Then use of white balance method for color tone correction,and get the final to be fog image.Finally,the brightest value in the local area of the fog image is the atmospheric light value,and dark color priori algorithm improved by the local adaptive template is used to estimate the transmittance and realize the fog.The experimental results show that the algorithm is effective and robust for nocturnal image fogging.