Image Dehazing Algorithm Using Single Image

In bad weather,because there are many suspended particles in the atmosphere,captured images generally suffer from non-trivial degradation which leads to obscure details and color distortion.Restoring the true scene appearance from hazy image,namely image dehazing,has the very vital significance in the field of practical engineering applications and image processing.The research works of this thesis consists of the following two aspects:(1):Based on local smooth assumption and local contrast observation,this thesis put forward a new algorithm.Holder coefficient is firstly introduced to the field of image dehazing in order to measuring local visible edges and local feature.According to Holder coefficient characteristics,we propose haze distribution model and removal haze model.In solving haze distribution map,rough map is firstly estimated by directly using Holder coefficient,then optimized by minimizing haze optimization function.The map has two jobs:1)Estimating Atmosphere Light Constant.2)Controlling the enhancement degree of visible edges.Removal haze model have two terms:Holder Coefficient term which measure visible edges and dense haze distribution map term which controls the enhancement degree of visible edges.Our core idea of removal haze is:Due to losing many visible edges at dense hazy region,Holder coefficient will be substantially increased in order to restoring the true appearance;Due to losing a small number of visible edges at thin hazy region,the enhancement degree of Holder coefficient will be suppressed in order to avoiding excessively enhancing image.(2):Our second removal haze model combining local model and non-local model.We systematically analyze these models,then find:1):At every non-local hazy line,there are some unstable points which cause transmission estimation error.2):During the derivation of dark channel prior,letting all dark channel equal to zero will lead to low estimated transmission.In order to solving these problems,our model combining local model and non-local model can get more accurate estimator and make removal image close to reality scene.The experience results show the first dehazing algorithm based on Holder coefficient have a better performance when the image includes more detailed information.However,the second algorithm combining local model and non-local have a better performance when the image lack detail.All in all,our algorim perform similar,even better than other dehazing algorithm.