Research On Image Haze Removal Algorithm Based Dark Channel Prior

The serious hazy weather has occurred repeatedly in many places of China in recentyears.According to figures released by the China meteorological administration,China’s averagehazy days are2.3days longer than the same period of ordinary year in2013,and has the mosthazy weather since1961;the hazy weather doesn’t reduce in2014,futher more,there is a trendtoward increase gradually.Fog is the aerosol systerm,which is composed by a large number ofsmall water droplets or ice crystals that suspend in the air near the ground,it is the product ofwater vapor condensation in the air near the ground;haze is composed by the particle in the airsuch as dust、sulfuric acid、nitric acid、organic hydrocarbon and so on.Fog and haze can loweropacity of the air,and blur vision,resulting in the decrease of visibility and a series of badeffects.Visible,hazy weather has caused serious adverse effects for people’s daily life.The formation of foggy image is influenced by many factors,and among them the mostimportant reasons include:large radius of particle ingredients in the atmosphere reduce thetransparency of the air,and lead to the reduced target visibility;at the same time,the sunlightwhich is scattered by the large radius of particle is enhanced.Under the common role of theabove two reasons,the quality of the foggy image is substantially declined.In order to obtain highquality dehazed image,we combined with the reasons of the formation of foggy image,and itsformation model,and process the foggy image.The main work includes:Firstly,we separate the sky area from the image.Dark channel prior is not applied to the skyarea,so we should to separate the sky area from the image if there is any fog in the skyarea.Furthermore,when processing the foggy image,we should obtain the atmospheric opticalvalue from the area where concentrates the most of the fog,namely the sky area of the image.Dueto the sky area’s characteristic, such as color and location,we obtain binary image at first,thenmade a series of morphological processing on it.Finally, we separate the sky area according to itscharacteristic,which is the sky area that often on the image’s upper part.Secondly, judge the foggy image and establish the foggy image library.To process the foggyimage,we should judge if there is any fog in the to be processed image.If the to be processed image is a foggy image, we need to process the image for image dehazing. Acoording to the darkchannel prior judge if the image is a foggy image,avoiding processing the no fogimage.Processing these images not only has no meaning, but also may occur the oppositeeffect.Depending on different scene,we divide the images into four classes:natural landscape、personage、vehicle especially automobile and urban architecture.Thus, we set up the foggy imagelibrary,for ease of the algorithm’s research and implementation.Thirdly,obtain image’s atmospheric optical value and transmittance.We select the maximumof the sky area in the image as the atmospheric value.Through analyzing,compared theconnection of the transmittivity、gray-scale image and inverse gray-scale image,we processedgray-scale image and inverse gray-scale image with median filter, then fuse them by imagefusion which is based on Contourlet transform,and treat the blending image as optimizedtransmittivity.Furthermore,according to the characteristics of gray-scale image and inversegray-sacle image,when the objective dehazed region is only the close or the distant view,we treatthe gray-scale image and inverse gray-scale image with median filter processing as optimizedtransmittivity, and obtain the expected result,and at the same time,the running time of thealgorithm could be greatly saved.Finally, according to the atmospheric optical value and thetransmittivity,we obtain the final dehazed image with the help of the foggy image formationmodel.Fourthly,get the subjective and objective evaluation index of the dehazed image.We need toevaluate the quality of the dehazed image.Subjective evaluation index conforms to human visualperception,but due to the various factors such as environment、person’s own mood and soon,subjective evaluation index can’t necessarily reflect the actual quality of the image.Commonobjective evaluation indicators include:mean square error、signal to noise ratio、informationentropy、average gradient、ratio of the added visible side、saturation percentage of black pixels、structural similarity and so on.These objective evaluation indicators reflect the quality of theimage partly,but due to the disparity with the human visual systerm,the result is not fully reflectthe image’s quality,and sometimes even contrary to the truth.According to the characteristics ofthe subjective and objective evaluation index,we evaluate the dehazed image by a methodcombined with subjective and objective evaluation index.Moreover,algorithm’s running time isalso an important index of evaluation algorithm,there would have no pratical if running time is too long,there would save time and resources and have real-time property if running time isshort.We saved algorithm’s running time and improved algorithm’s practical value on thepremise of ensuring dehazed image’s quality.Fifthly,apply of image haze removal algorithm based dark channel prior.License plateof the images that have no fog can be identified well,but license plate of foggy i-mageis hardly due to the low contrast and clarity.So we applied the algorithm that we suggested to license plate recognition field,dehazed the license plate area and reduced thedifficulty of identify.We used MATLAB7.1software development,and operational procedure on the personalcomputer of Pentium(R) Dual-Core、2.59GHz、3.00GB.We processed foggy image by i-mage haze removal algorithm based dark channel prior,and obtained high quality dehaze-d image with a relatively short time,increasing the practical value of the image.We comp-ared and analyzed each objective evaluation index of dehazed image,and got a good exp-ected effect.We applied the algorithm in the license plate recognition field,and got a hig-h recognition rate.