Single Haze Image Recovery Using Dark Channel Prior

The images captured in the foggy weather conditions will be severelydegraded because of atmospheric light scattering. The specific performance ofthe degrad can be described as declining in image contrast, lower resolution ofthe shooted scene, bleaking of true color images and so on. Thesecharacteristics result in greatly reduced the value of the image. Sometimes,even if there is no fog, tiny particles in the air (for example: dust, fog andsmoke) caused by atmospheric light scattering will lead to image degradation.people’s production and life is impact a lot by the images degraded byfog.Therefore, a wide variety of defogging algorithm came into being.This paper is to study a methord to remove haze in a single image usingdark channel prior.The atmospheric scattering model provides a basealgorithm to this methord.The mainly research includes as following:(1) Conclude and study of previous algorithm of haze removel. Make adescription of the basis of the method used by the paper, and make a simpledescription of digital image processing and atmospheric scattering model. Wecan get the dark channel by the following steps,firstly choose the minimumvalue in color sapace of each pixel,secondly,take the minimum value in a localregion of each pixel.The atmospheric light intensity can be estimated in the darkchannel.We choose some maxmum gray intensity pixel in the darkchannel as theseeds.The corresponding pixels valeu in the original channel is the atmospheric lightintensity estimation. Using dark channel prior conditions, and the atmosphere lightintensity estimation, combined with atmospheric scattering model, we can estimatethe depth of the scene, recovery the haze-image.(2) Make a implement of previous theoretical method and analyse theshortcomings. Making some improvements to the previous method embarkedon transmittance optimization direction.In the prcess of researching,we found that this method has some shortcomings.First,the dark channel prior is notsuitable in some regions where the pixels have powerful gray intensity in everyband.Second,the efficiency of computing the transmittance optimization.Weavoid the first shortcoming by computing the transmittance partable. In brightregion we set a tolerance threshold, when the pixel tolerance is less than athreshold,we see the transmittance estimation and actual value is relatively large,and need to be corrected. If the tolerance is greater than a threshold, thetransmittance is accurate, and do not need to be corrected. Aiming at the problem oflow efficiency of transmissivity, we use the method of Guided Filtering to replace thesoft matting optimum transmittance. Guided Filtering filter is the improvement ofthe soft matting, its essence is a kind of bilateral filtering.This methord. needs a bootimage as the guidance, through a series of processing to achieve input image andboot image is quite similar in visual effects.(3) Using MATLAB and C++mixed programming to carry on thealgorithm.We speed up the process by more than twice by using c++to callmexw32.This calling make a full use of the powerful Matlab matrix arithmeticfunctions and mathematical function library and avoid the long time spendingon loading the module that is not requested in the algorithm.