Algorithm Of Image Dehaze Based On Dark Channel Prior

Computer vision system has been widely used in all kinds of outdoor scene,such as the urban traffic,video monitoring,remote sensing imaging,aerial,robot systems,industrial control,etc.And it is playing more and more important role in modern life.However,in the fog,haze,smoke,dust and other harsh weather conditions,due to the scattering effect of the large number of tiny drops of water and aerosol particles floating in the air,the images required by the computer vision system have low contrast,and they lack in color information,and they also have too much noise.This could seriously degrade the its performance in outdoors situations.So,restoring the foggy images is of vital importance.This article conducts a study about this topic.(1)We have conducted a further study about the physical imaging model of foggy images,and we also did a deep analysis about the algorithm of single image haze removal using dark channel prior,then we could know what is its advantage and disadvantage,finally,we make an improvement on the existed algorithms.(2)According to the characteristics that the fog's influence on image quality is mainly focused on the low-frequency component,image dehazing is simplified into low-frequency image dehazing using the wavelet transform,this could greatly improve the computational efficiency,and also improve the practical value.(3)For the defect that the dark channel prior will not work in the sky region,we separate the image into sky region and non-sky region using image segmentation methods,and refine the transmission in sky region,and get better airlight value using pixel values in sky region.(4)In the process of refining the corase transmission,the soft matting algorithm cost a lot of time and memory space,this greatly decrased the performance of the dahaze algorithm,so,in this thesis,we use guided filtering instead of soft matting algorithm.(5)The color of the restored image is a bit darker,so,we use the CLANE algorithm to process the images,and introducing a reasonable evaluation index to evaluate the quality of the restored images.