Research On Fog Image Processing Methods Based On The MSR

With the rapid development of Information Age,Computer Technology is gradually fused to Image Multimedia Technique,this integration is closely related to our daily life.Fog weather is a common natural phenomenon,the acquired picture information is missing,although the human eye can basically distinguish the scene,the visual effect is not ideal.Thus,it is very important to use the image processing technology to achieve the effect of removing haze,get a clear image and reduce the safety hazards caused by fog weather.The traditional algorithm based on histogram equalization fog image processing and direct Multi-scale Retinex algorithm can reduce the fog interference,but image enhancement is uneven after processing,the detail information is not clear and the color fidelity is low.In summary,it is theoretical and practical to improve the traditional fog image processing method.The main contents and achievements of this thesis are as follows:(1)Research on the traditional fog image processing method,this paper introduces three traditional algorithms to remove haze,which are histogram equalization algorithm,homomorphic filtering algorithm and multi-scale Retinex algorithm,and simulate the basic algorithm.Through the simulation found that the local detail enhancement is not obvious,overrall enhancement uneven and color fidelity is low when removing the fog.In this paper,an improved haze image processing method based on multi-scale Retinex algorithm is proposed for the shortcomings of traditional fog image processing methods.(2)Aiming at the defect of the algorithm of histogram equalization of fog images,such as uneven image enhancement,high-frequency component reduction and unclear detail information,model one is proposed to solve the defects of histogram equalization algorithm.Firstly,according to the principle of Retinex algorithm,the fog image is decomposed into the incident component and the reflected component,then weighted average operation of the incident and the reflected components is taken as the maximum value of the absolute value,finally,to merge the incident component and the reflected component.It is found that the high frequency components of the image are increased and the dynamic range is increased,and the haze effect is good.(3)Aiming at the defect of the direct Multi-scale Retinex algorithm,model two is proposed for the problem that the color fidelity is poor,the image is gray and the detail is not obvious.The algorithm first reverses the original image,then decomposes the original image and the inverted image into RGB three channels and extracts the information of the high frequency components,after linearing stretch and powering transformation of high frequency components,Then the RGB layers are merged to obtain the processed original image and the processed inverted image respectively which is merging with linear proportional relations.It is found that the color of fog image and the scene in the original image is restored.At the same time,the high frequency information is enhanced,the effect of the image is more obvious,and the haze effect is good.