Optimizing Technology Of Dark Channel Prior Dehazing Based On Sky Region Segmentation

Recently,due to air pollution and other reasons,fog or haze and other unsatisfactory weather appears frequently in large area.The images acquired by outdoor imaging equipments usually have a seriously bad quality with a lot of detail information lost.This will directly restrict and influence people's daily activities.Therefore,in order to enhance the visual effects of the image and facilitate the subsequent processing of computer vision,it is necessary to remove the haze in hazy images.This paper thoroughly studied and analyzed the problems in dark channel prior(DCP)dehazing algorithm.The transmission values in sky region where DCP law is failed will be smaller than actual values after estimation,and this will lead to obvious noise amplification and color distortion in sky region.Besides,the transmission estimated values near the edges of depth changes in non-sky region will be inaccurate,and this will lead to unsatisfactory dehazing effects caused by the attenuation of dehazing intensity at these positions.So this paper will firstly extract the sky region in hazy images with sky region segmentation to deal with the problems in these two regions respectively.Then the transmission estimation in sky and non-sky region will be optimized and improved separately.After that,some further visual optimizing methods will be used on the whole dehazing effects to make the dehazing visual effects better.To solve the problems in sky region,a method combining K-means clustering with enhanced edge extraction is proposed for sky region segmentation.First,the candidate region is extracted with proposed judge value by K-means clustering.Next,the sky connected region is extracted by enhancing edges.Then this two results can be combined by using and operation.And after some improvements on the selection strategy of two results,more perfect and practical sky region segmentation result can be obtained.Then the transmission values in sky region of the hazy image will be adaptively corrected in a certain way based on the changes of original values.After that,better dehazing visual effects will be acquired.To solve the problems in non-sky region,a method called block-to-pixel interpolation will be used to improve the processing of calculating the DCP images.This will solve the problems caused by the inaccuracy of transmission values near the edges of depth changes to some extent.To further optimize the visual effects in dehazed images,the optimizing and enhancing methods of increasing the stability of atmospheric light estimation,promoting the definition of dehazed images and adjusting the tone of dehazed images will be proposed and used.Then better dehazing visual effects will be obtained eventually.Finally,the proposed improved dehazing algorithm has been programmed to be realized.The whole algorithm has been accomplished and the experiments have been performed on personal computer using VS2010 and OpenCV2.The resulting dehazed images in experiments have been analyzed on subjective and objective level,and the authenticity and validity of this improved algorithm has been proved.