The Research Of Single Image Haze Removal Algorithm

In the fog and haze weather condition,due to the scattering and absorption of large number of suspended particles suspended in the atmosphere,making contrast decreased and color attenuated of the outdoor image captured,so it is difficult to distinguish the characteristics of the object,and the viewing of image is lowered,affecting the subsequent processing of the image.Therefore,in the application of the computer vision direction such as target tracking,video monitoring and remote sensing,the research of image haze removal technology has great significance.With the development and mature of related computer technology,the research of image haze removal technology also has become the hotspot issue.The haze removal algorithm is mainly divided into image enhancement and image restoration based on physical model.Two types of algorithms have their own advantages and disadvantages,but in recent years,the haze removal algorithm based on physical model has become more common.Among them,the haze removal algorithm based on median filtering has received widely attention because of faster processing speed and better processing result.However,on the one hand,the atmospheric veil obtained by median filtering lost a lot of image edge information,which can not reflect the real scene depth information,leading to haze removal is not complete;on the other hand,the transmission estimation value of sky area is low,resulting in color distortion and noise in the image processing result.So,in this thesis,we mainly study the haze removal algorithm from two aspects,including the veil and the transmission.Firstly,a single image haze removal algorithm based on weighted guided filter is proposed.Based on the guided filter,the Canny operator is used to adaptively set the weight of edge region and the flat area in the input image,so that the output image retains more edge information and smoothes more noise and texture information.According to this feature,replacing the median filtering with the weighted guided filter to obtain the atmospheric veil which contains more rich edge information,then replacing white balance operation with dark channel prior to estimate more accurate atmospheric light.The experimental results show that the algorithm has a faster processing speed,and the haze removal of edge is more complete.Secondly,a single image haze removal algorithm based on the atmospheric veil optimization and transmission correction is proposed.On the one hand,the atmospheric veil algorithm can not accurately reflect the depth information of the scene because of the introduction of noise and texture information.Whether using the median filtering or the weighted guided filter,the restored image has incomplete haze removal.On the other hand,the transmission estimation value is low in the sky,resulting in noise and color distortion in the restored image.So,firstly,extracting the sky area by using threshold segmentation,and then reducing the gray value of the corresponding region of the minimum image of each color channel appropriately.Secondly,obtaining the optimized atmosphere veil through two weighted guided filter operations of different guidance images.At the same time,we use dark channel prior to estimate the atmospheric light.Finally,the restored image is obtained from the atmospheric scattering model.Experimental results show that the proposed algorithm has a more complete haze removal effect in discontinuous areas of the depth of the scene,while the sky area avoids noise and color distortion.Also,the white area similar to the atmospheric light avoids the dark color.