Research On Single Image Dehazing Algorithm And Effect Assessment

In recent years,with the rapid development of China's economy and the acceleration of China's industrialization,the environmental pollution of the city is becoming more and more serious,especially the frequent occurrence of haze.Not only brings inconvenience to people's travel,but also brings severe challenges to the field of urban road traffic monitoring,satellite remote sensing and military reconnaissance.The appearance of haze causes the quality of monitoring equipment to be degraded seriously,which will often increase the difficulty of feature extraction and target tracking.Therefore,the research of image dehaze algorithm has become a hotspot in the field of computer vision in recent years,which has the characteristics of adapting to the development of the times and wide application prospect,and it is worthy of domestic and foreign scholars' research and development.This paper mainly focuses on the following three aspects: haze image imaging principle,haze image dehaze algorithm and dehaze algorithm performance evaluation.The main research contents are as follows:(1)The study of haze image defogging algorithm involves many subject areas.Firstly,the formation mechanism of haze,the theory of atmospheric scattering and the model of haze image degeneration are studied,on the basis of which the imaging principle of haze image,the image modeling and the characteristic analysis of haze image are studied.(2)Study the dehaze algorithm based on image enhancement,the histogram equalization algorithm mainly improves the haze image gray level distribution concentration problem,thus enhances the contrast;the homomorphic filter fog algorithm is processed in the image frequency domain,enhances the high frequency component and suppresses the low frequency component,not only enhances the contrast,but also enhances the image edge information.The Retinex algorithm is processed in the image frequency domain,the dehaze effect is better than the homomorphic filtering algorithm,and the four improved Retinex algorithms are compared and analyzed,and the optimal algorithm for fog removal is obtained.(3)Study the dehaze algorithm based on dark channel priori,which has obvious advantages in the effect of dehaze algorithm,but the algorithm uses Soft Matting to optimize the rough transmittance resulting in higher algorithm complexity;this strategy uses the Guided Image Filtering instead of Soft Matting graph to reduce the algorithm complexity,but the effect of fog removal is a little worse.Aiming at this shortcoming,this paper proposes an optimized dehaze algorithm,which combines the Frankle-mccann Retinex algorithm with the Guide filter algorithm to optimize the rough transmittance,and improves the principle of the original algorithm's atmospheric light estimation and filter window parameters selection,and compares the simulation of the fog algorithm with the result analysis,The advantages of the improved algorithm in this paper are verified.(4)In the last part of this paper,we mainly study the method of image evaluation,and finish the image evaluation of the proposed algorithm,and verify the advantage of the improved algorithm by the result analysis,and find that the improved algorithm is more suitable for the image with larger haze density.