Research On Image Dehazing Method Based On Dark Channel Prior

Haze limits the visibility of outdoor scenes and reduces the image contrast.Haze removal can improve the sharpness and contrast of images,restore clear and realistic scenes by reducing or eliminating haze interference in images.In this thesis,we propose some improved ideas and algorithms about the dark channel prior theory.The main influencing factors such as atmospheric light value,transmission,and dark channel are researched and some proved effective methods verified.The main research contents and innovations:(1)The inaccurate determination of the atmospheric light value may cause the problem of color shift in the restored image when the dark channel dehazing method is applied.To solve this problem,we propose a novel method for estimating the atmospheric light value based on binary search idea.The analysis of the gray values' mean-variance feature of image pixels in the sky region is the key procedure.First,determine the part that contains the sky area according to the characteristics of high gray value of the sky area pixels in the image.Second,locate the sky area once again according to the feature that the grayscale variance of the pixels in the sky area is small,determine the sub-regions of the image that contains the sky part.At last,the pixel average of the sub-regions in the sky is taken as the atmospheric light value of the image.Experiment results show that this method can quickly identify the sky area in the image,get reliable estimates of atmospheric light,and then improve the color shift in the dehazed image.(2)The inaccurate calculation of the transmission at the position where the depth of field changes sharply in the image can cause the halo phenomenon easily in the dehazed image with the dark channel prior method.We propose a transmission optimization method based on Gaussian curvature filtering.Gaussian curvature filtering has enhanced smoothing capability and can protect the edges of the image better by utilizing the discreteness of the image and the continuity of the differential geometry.First,the coarse transmission is obtained according to the dark channel prior theory.Then,the Gaussian curvature filtering is applied to obtain an optimized transmission.Finally,the haze image is restored according to the degradation model.Experiment results show that it can effectively suppress halo phenomenon and the color shift phenomenon in the restored image,and can retain more detailed information.(3)In some parts of the image,the depth of field may change drastically.The fundamental cause of the halo phenomenon is the deviation of dark channel values.This can lead to the inaccurate transmission..We propose a halo effect removal method based on image dark channel correction.First,compute the minimum image and the dark channel image corresponding the haze image.Second,we can get an adaptive threshold with the difference between the gray values of the minimum image and the dark channel image.With the threshold value,the positions where the depths of field change drastically(usually the edge position of the object in the image)can be determined.So,we can obtain a more accurate dark channel value.The method can improve the execution efficiency of the entire algorithm because it eliminates the optimizing procedure of the transmission.Experiment results show that it can effectively suppress the halo effect and improve the dehazing effect.And the overall image quality can be enhanced.