Study On Color Correction And Defogging Algorithm Of Foggy Images At Night

This paper mainly studies the color correction of foggy images at night and the algorithm of fog removal.Most of the models used in existing defogging methods are designed to describe daytime smog.The daytime model assumes that a single uniform light color is attributable to a light source that is not directly visible in the scene.However,night scenes often include visible light sources of different colors.These light sources also often introduce obvious glow that does not exist in the haze in the daytime.Therefore,the image of haze at night may be affected by uneven illumination of artificial light sources,which has problems such as low overall brightness,uneven illumination,serious color bias and high noise,etc.,making it more difficult to remove fog.Aiming at these problems,this paper makes an in-depth study on the image restoration algorithm in foggy night.The main work of this paper includes the following two aspects:Firstly,for the serious color distortion of haze image at night,the basic principle of automatic color equalization(ACE)and the fast approximation method are described,and a color correction algorithm of ACE which can suppress halo is proposed.ACE has proven to perform effective color constancy correction and satisfactory tonal reproduction,converting color correction images from RGB color space to Lab color space.Then,we used the L-channel in the color space to adjust the illumination using the gamma-function improved Contrast Restricted Adaptive Histogram Equalization(CLAHE).The experimental results showed that this method could effectively correct the color bias and suppress the halo enlargement caused by the processing.Secondly,this paper introduces two algorithms for removing fog in night images.One is the algorithm for removing fog in night haze images based on super pixels.Firstly,a method based on superpixel is introduced to calculate the atmospheric light and dark channel value of each pixel in the night foggy image.The transmission image is decomposed from the dark channel of haze image through weighted guided image filter.Because superpixels are usually well attached to the boundaries of objects,small local window sizes can be selected,and the details of fine-structure areas are better preserved.In order to avoid obvious noise in the sky area,an adaptive threshold is added to the transmission image when the image of haze at night is restored.Compared with existing methods,this algorithm can remove haze better,and local fine details are well preserved.Combined with the color correction algorithm proposed in this paper,the final night fog-free image is closer to the natural scene,and the problem of image distortion caused by color bias is solved.In addition a kind of effective image fusion method is proposed to enhance the fog scene at night is,the algorithm based on multi-scale fusion method,including underexposed and enhance image contrast of multipath input to perform improved multi-scale fusion,the two groups of input image with color correction methods in addition to color,and their corresponding weights derived from each ratio,different input respectively from underexposure and contrast limited adaptive histogram equalization processing,finally,using the Laplacian pyramid decomposition in multi-scale mixing export input and normalized weighted graph.Experimental results show that our method is effective in terms of computational efficiency and output quality compared with the existing techniques.