Research On Image Dehazing Algorithm Based On Light Veil Correction And Feature Extraction

With the rapid development of science and technology,computer vision is widely used as a hot area of artificial intelligence in video surveillance,autonomous driving,medical imaging and other areas.However,they are susceptible to the extreme weather and other uncertainties that can lead to degradation or even failure of the system,for some computer vision systems installed outdoors.In adverse weather conditions such as haze and dust,suspended particles in the air can scatter and refract the light reaching the imaging device,resulting in poor visibility,unclear subject targets and of lower color fidelity in the images captured by the computer vision system.This has a detrimental effect on subsequent work.Therefore,it is a key pre-processing step in vision tasks to eliminate the effects of haze on the imaging process by certain means and obtain a high-quality image.Based on this,this paper presents a set of new and improved algorithms with correction of the physical degradation model and haze feature extraction,based on the atmospheric scattering model.Through in-depth analysis and study of various existing classical algorithms,considering the mainstream algorithm technology line,to address their shortcomings.(1)To address the problems that the dehazing algorithm is prone to distortion in highlighted region of an image and edge pixel loss from the minimum filter,a dehazing algorithm with an atmospheric veil correction model under regional detection is proposed.First,the initial atmospheric light veil is obtained by using the improved dark channel,and the atmospheric light veil correction model is established for it.Then,the color distance and color distance histogram fitting curve of the haze image are used to pixel-wise detect the region where the initial atmospheric light veil is corrected.Combine corrected region of an image and color distance to construct the correction term.Finally,the atmospheric light is estimated by using image information entropy as the objective function of the quad-tree subdivision algorithm.It is proved that the proposed method effectively solves the distorted problem in the highlighted region and the block artifacts produced by minimum filtering,and the recover image is clear and natural,with rich detailed information and thorough dehazing effect.(2)Due to the influence of medium particles in the air under hazy weather,images captured by imaging devices usually suffer from low contrast and color loss.To address these problems,an atmospheric light veil estimation dehazing algorithm based on haze distribution is proposed in this paper.First,the initial haze distribution image extracted by twelve features(such as image information entropy,HSV spatial saturation component,yellow and blue components,image sharpness,etc.)of the image is subjected to threshold and refinement to obtain more accurate haze distribution information.Secondly,the atmospheric light veil estimation model is established by analyzing relationships among the depth of field,haze concentration,and atmospheric light veil.Finally,the atmospheric light region is selected by luminance weighting map and adaptive atmospheric light threshold,and then recover the degraded scene.Through experimental proof and result analysis,this algorithm restores images clearly and can retain detailed information in the image.