The Study On Single Image Dehazing Based On AS Model

Because of the rapid development of the information science technology,the application of the visual system in the capture and processing of scenes is used more and more widely.However due to the effects of tiny particles in the air,images captured have many problems,such as low clarity,contrast and color changes,many scenes of images indistinction,which not only affect the image visual effect,but also cause great inconvenience to the subsequent information collection and processing.Therefore,it is significant to study the dehazing technology,which has wide range of application.At present,there are two kinds of dehazing technology to color degraded images:first one is emphasizing global or local contrast,without considering causes;second one is the images recover technique based on the atmospheric scattering model.This paper belongs to the second one,starting from the physical causes of image degradation,expounds on AS model,and reviews some classical AS model-based dehazing methods.This paper studies single image dehazing algorithm based on Atmospheric Scattering(AS) Model,the specific work are as follows:(1)According to the characteristics of atmospheric light in the atmospheric scattering model and image color space,this paper uses quad-tree hierarchical searching method in YCbCr space,which can avoid estimating unreliable atmospheric light value;(2)Becausing of the local smoothing of transfer rate, assuming that local transfer rate is same in each subclass,AP clustering algorithm,fast and stable,is used to instead of rectangular window which needs to be set artificially,divide degraded image into several subclasses adaptively,to estimate local transfer rate;(3)in general,a hazy image exhibits low contrast,so this paper estimates local transfer rate to enhance its contrast. However, the over compensation of the degraded contrast may cause pixel values overflow and information loss. Therefore, this paper formulates a cost function that consists of the contrast term and the information loss term and estimates the optimal transfer rate value for each subclass by minimizing the cost function;(4)Guide filtering algorithm can keep edge details of the image,this paper carries on the improvement on running speed and uses smooth window to keep edges,highlights the edges and details information of the image.The experimental results show that the proposed method can improve the visual effect of image more effectively,from the objective angle evaluation,our method is superior to other three classical methods in performance.