Research On Foggy Image Defogging Method

With the advancement of modern technology,the field of machine vision has seen unprecedented growth.Images and videos as a source of visually accessible information play an important role.However,the quality of images captured in daily life are affected by many external weather conditions.Fog is a common factor.Images captured in foggy conditions suffer from reduced contrast and loss of information,appearing grey and blurred,affecting traffic enforcement,drone photography,military defense and so on,making image defogging techniques particularly important.Among them,the methods based on image restoration have been widely researched by the academic community,with remarkable defogging effects.However,there are still shortcomings such as color distortion,incomplete defogging and discomfort to the sky area.This thesis deeply analyzes and explores the theory and shortcomings of classical de-fog algorithms and improves them.1.Analyze the formation process,characteristics and effects of fog.Study the physical model of degraded image formation in foggy environment.Explore the classic theory of dark channel prior and analyze its rationality and deficiencies.Then improve the defogging effort from the root cause in combination with the physical model.Study the image quality evaluation method,and establish indicators to compare the recovery effects of different algorithms.2.Aiming at the shortcomings of using a fixed window to calculate the dark channel,the risk of misjudging the foreground of image in calculating the atmospheric light value,and the inaccuracy of the initial transmittance calculation in the defogging method of dark channel prior,an improved method is proposed.Firstly,acquire the dark channel of the foggy image.Change the size of the filter window from fixed to adaptive to eliminate the block phenomenon in the dark channel map.Secondly,correction of the atmospheric light value.The combination of the bright channel and the adaptive dark channel is adopted to fuse the atmospheric light value by setting weight to atmospheric light value of two channels.Thirdly,optimize the transmittance.The initial transmittance is obtained using the dark channel map obtained from the adaptive window,and optimize it using the guided filter,regarding the gray image of the original image as the guided image.Finally,the image imaging model is used to obtain the recovered image.3.Aiming at the shortcomings of dark channel theory on the failure of sky area,an image defogging method based on sky area segmentation is proposed.For an image containing the sky area,segmentation is the first step.The Canny operator is used to obtain accurate edge information.And the Otsu algorithm is improved to select the best threshold,and the binary image segmented is obtained.In order to fill the gap in the foreground of the binary image,the morphological closed operation is adopted.Then an accurate segmented image is obtained.Secondly,since the actual value of the atmospheric light value should be at infinity in the image,it is more accurate to obtain the value in the sky area.Thirdly,for the estimation of the transmittance,obtain the transmittance of the sky area and the non-sky area respectively and set weight for them by using the binary image as the weight map.Then the guide filter is used to refine it.Finally,the clear image is obtained.After comparative experiments with other algorithms,evaluate and analyze the image quality from subjective and objective level,the recovered images of the two improved algorithms in this thesis have richer detail information,smoother edges,more realistic and natural visual sense.