Research On Dehazing Algorithm For Single Image

In recent decades,the industrial economy has achieved a leap-forward development,but followed by the continuous deterioration of the ecological environment,among which frequent haze weather seriously affects people's quality of life.In the haze weather,a large number of atmospheric particles,such as small water droplets and particles,which are suspended in the air,affect air quality as well as outdoor imaging equipment.These atmospheric particles will scatter the light to varying degrees and make it deviate from the original propagation path.Therefore,the images taken under the haze weather will have problems such as reduced contrast,color shift,and poor visibility,which significantly reduces the image quality and is extremely unfavorable for engineering applications such as remote sensing satellite shooting,drone aerial photography,and autonomous driving.Therefore,it is extremely necessary to conduct research on how to remove haze from images taken in haze weather.This thesis studies the dehazing algorithm for a single image,and the main work and innovation are summarized as follows:1.Aiming at the poor performance that the dark channel dehazing algorithm in sky region and white areas,this thesis makes use of the advantages of color attenuation prior and dark channel prior in different areas,proposes a weight mapping function based on dark channel,and uses the weight map obtained by mapping to fuse the media transmission map obtained from these two prior methods.At the same time,in order to improve the fusion accuracy,the Laplace image pyramid is used.The atmospheric light value is determined by using the quad tree to search the hazy image in a hierarchical manner.The comparative experiments show that the proposed method can remove the haze very well,and the objective evaluation index can also show that the proposed scheme has certain advantages.2.Aiming at the problem that color attenuation prior dehazing is not thorough enough,this thesis proposes a dehazing model based on the depth estimation of hazy images from the perspective of model construction.Based on the relationship between dark channel in the dark channel prior and fog concentration,a log dark channel with depth correlation of hazy image is proposed,and the derivation of the relationship between log dark channel and hazy image depth is given.In addition to using the original S component and V component in the color attenuation prior model as the depth feature of the hazy image,the square of their difference and the local contrast of the image are also used as features.These features are used to construct the hazy image depth estimation model,and then the model parameters are obtained by regression training.In addition,the two atmospheric light value estimation schemes in the dark channel prior and the color attenuation prior are jointly used to estimate the atmospheric light value.Finally,the proposed method is proved that it can remove the haze effectively by the experimental results,and the sky area is naturally dehazed and not over-enhanced.The proposed method is persuasive in both subjective and objective evaluation,which shows that this method is effective.