Single Image Dehazing Removal Based On Concentration Scale Prior

With the increasing of human activities and the continuous decline of air quality,the density of particles in the air and the degree of refraction and scattering of global atmospheric light increases are both increased.The fog and haze occur from time to time,which results in the visual information captured by human eyes or imaging equipment being affected.There are some phenomena such as blurred edge and lost details in image,which have a great impact on human production and life.At present,there are many kinds of computer vision image processing algorithms,which can restore the clarity of the image and eliminate the damage of fog to the image quality.However,the restoration effect of hazy image definition needs to be improved.This thesis proposes a novel and effective single image haze removal based on concentration scale prior,and achieves the expected defogging effect.Firstly,this thesis introduces the development of image dehazing at home and abroad from two aspects: the traditional method based on physical model of atmospheric scattering and the method based on deep learning neural network.Secondly,by describing the principle of atmospheric scattering model in detail,we can understand the essence of fog image imaging.In order to solve the problem of single image defogging,we need to make some assumptions about the model.Finally,it introduces the classical dehazing model,dark channel prior and color attenuation prior.This thesis proposes single image dehazing model based on concentration scale prior,it is an improvement of gamma transformation.In order to recover the best haze free image after gamma transformation,this thesis uses the average value of the difference between brightness and saturation as the fog concentration scale.According to the different concentration scales obtained from different images,the reasonable parameters are estimated as the index of gamma transformation,so that the dehazing effect of a single image is the best.The model has pertinence and accuracy for the fog concentration of the image with fog.From the comparative experimental results,our model has the largest PSNR and SSIM values and the shortest operation time,whether in the synthetic or in the real-world hazy image.Next,the single image haze removal based on concentration scale prior is improved in two aspects.The first one is to improve the fitting method of gamma estimation function;the second one is to redefine the optimal gamma value.Both of the improved methods have obtained the optimized experimental results.