Research On Clearness Model And Application For Haze Image

For an image or videos captured under the fog weather,due to the influence of haze weather,the contrast and sharpness are delcined,which makes the effective information reduced in the image and will seriously affect the application of the outdoor vision system in human life.Therefore,it has a high research value for modeling for clearing foggy images and recovering the effective information of images.In this thesis,we mainly research on clearness model and application for haze image.The research contents are as follows:1.The method of image dehazing modeling based on atmospheric scattering model is expounded,and some mainstream dehazing algorithms are explained in detail.We evaluate these algorithm by subjective and objective image quality evaluation methods.Subjective image quality evaluation methods are used for natural images,and objective quality evaluation of these algorithms are evulated with O-Hazy data sets.2.By comparing the state-of-the-art haze removal algorithms,we found that the dehazing results of the non-local image dehazing algorithm is more prominent,but the algorithm is easy to cause the existence of haze shadow around the object,and due to the lack of time information,it is easy to cause flickering artifacts in the video dehazing by directly applying this algorithm to the each video frame.For these defects,we propose a non-local video dehazing algorithm based on haze line.This algorithm mainly improves the regularization optimization scheme for the medium transmission map,so that the medium transmission image's texture information is presented in the smooth process,and a adaptive smoothing factor is proposed,so that the optimization of each pixel of the medium transmission map can be adaptively smoothed,effectively avoiding the existence of haze shadow around the object.For the video,we use optical flow method to limit the medium transmission map,so that the recovered video image has good time coherent information and avoid the occurrence of flickering artifacts.3.For the unmanned aerial vehicle remote image,we propose a UAV remote sensing image dehazing model based on dark channel prior.We introduce the rationality of applying the dark channel prior principle to the UAV remote sensing image dehazing,and we use a quad-tree hierarchical searching method to estimate the atmospheric light value,this can effectively avoid the influence of white objects in the image on the estimation of atmospheric light values.In the process of refining the transmission map,we propose a new regularization optimization scheme,which enables the UAV remote sensing image captured in foggy days to be well recovered.