| With the development of computer technology, the information and intelligence equipment is gradually penetrated into all aspects of our social life, as an important branch of computer application technology, computer vision has been widely and deeply applied in many fields like traffic monitoring, vehicle navigation, driver assistance and aerial detection. However, the interference from bad weather is always an important factor of affecting the image quality. Especially in recent years, the dust and haze weather frequently appear in most areas of our country, which brings a lot of distraction to the normal operation of visual system and seriously affects the visual effects and post-processing of the images. Thus, the research about single image dehazing has become an important subject of current computer applications. The dehazing algorithm based on dark channel prior has obvious advantages in recovery effects and stability which makes it's used widely, but there are still many shortcomings. For example, the false color phenomenon in bright areas, the real-time of algorithm processing and the accuracy and versatility of model parameter estimation are not high enough, the restored image has color saturation phenomenon and darker overall visual effects. This thesis analyzes the latest developments of image dehazing technology, and combines with the application status of dark channel prior to propose some improvements, and further improve its practical value of the algorithm.Since the current image dehazing algorithms based on dark channel prior has many deficiencies, this thesis proposes two new parameter estimation algorithms.(1). A contrast restoration algorithm for single image based on Physicals model: on the basis of the dark channel prior method, the atmospheric scattering model is analyzed and then the fog distribution under the influence of dark channel image is summarized, according to which the outdoor images are added fog. The transmission is estimated through the field depth relationship between the reference image after adding fog and the outdoor image, then the purpose of defogging can be achieved.(2). Fast image dehazing algorithm based on relative transmission estimation: on the basis of the dark channel prior method, the relationship between field depth under haze condition and minimum image of color channel(RGB) images are analyzed, the preliminary estimation of transmission is achieved by relative amounts of depth and corrected by the characteristic of atmospheric light curtain. At last, the clear image can be recovered by the physicals model. In addition, in order to improve the overall visual effect of the restored image, the thesis presents two methods about brightness enhancement:(1). An enhancement method based on difference of RGB channel: this method takes the ratio between pixel value of every channel into account, and seeks an unify function to enhance the difference between RGB image and their mean image.(2). An enhancement method based on reference of gray image: at first, using the objective function to enhance the gray image of restored image, then determining the uniform enhancement coefficient by gray image and its enhanced results. Finally, using the coefficient to enhance the RGB channel image.Experimental results show that the proposed algorithms in this thesis take restoration quality and processing speed into consideration. For different haze conditions, the proposed algorithms can not only get higher quality restored image, but also can reduce the complexity greatly. The follow-up enhancement methods can effectively enhance the image's details in dark areas, and improve the overall visual effect and layering images.The achievements of this research are improvement to the current dehazing technology. Not only the image restoration technology research is deepened with high research value, but also the shortcomings in the practical application of the technology are overcome to improve the utility value of the dehazing system. |
PDF Full Text Request