Research On Image And Video Clearness Restoration Technology In Haze Weather

Computer vision systems are widely used in outdoor scenes work, such as video surveillance, urban transport, aerial photography, remote sensing imaging and so on, which has played a very important role in areas of daily life and national security. However, most of the existing outdoor visual system is particularly sensitive to the environment, which need clear and accurate contour extraction image details. Especially under conditions of haze days, collected images are scattered by atmospheric particles and severely attenuated, in which ambiguity and a decrease in contrast, color distortion and degradation exist. They directly limit the normal functioning of the visual system of outdoor utility. Therefore, in foggy conditions, how automatic, real-time images and eliminate the impact of fog on the scene target video frame has become a hot topic of image processing.This paper studies the foggy image and video clarity recovery technology, and introduces the basic theory of image and video algorithm used to fog, including atmospheric scattering physical model, dark channel prior theory and Retinex algorithm technology. On the basis of the existing traditional algorithm to study the fog analysis, we propose two methods for a single image to haze image defogging. For video haze remover, a fast video defogging method is proposed. specific research contents are as follows:(1) A single image defogging algorithm is studied. Images taken in foggy weather are seriously degraded due to the scattering of atmospheric particles. A simple and effective haze removal algorithm from a single image is proposed. First, a halo evaluator is designed to detect halo zone. Then, precise transmission rate is obtained by weight fusion of single pixel based rate and block area based one, both taking the prior of dark channel. The weight is determined according to the halo evaluator. Finally, a parameter is added for image recovery to limit the low transmission and to protect the sky area. Experiments show that compared with other methods, more vivid and natural images can be recovered by the proposed method, especially at the edges of the foreground and the background and in the sky area.(2) Aiming at images taken in foggy weather exist serious degraded phenomena further study such as low contrast, color distortion and obscure. A new haze removal algorithm from a single image is proposed. First, from the atmospheric physics model, the color balance correction of the image, for the image-based physical model to estimate the fog light halo effect occurs and Retinex algorithm based on non-physical model to fog is not complete, and the process appears in color distortion, physical model and Retinex algorithm are combined to estimate transmission which improves computation efficiency and prevents the halo of light and color distortion. Then, an adaptive weighted fusion method is proposed to calculate atmospheric light. The fusion weights are estimated via image dark and bright segmentation. The obtained atmosphere light value is more reasonable and accurate. Experiments show that the algorithm is simple and effective, and recovered images are clear and natural, achieving a good defogging effect.(3) Fast Video defogging algorithm is studied. Aiming at the single image defogging processing algorithms directly applied to the video frame by frame on the video player will slow and the adjacent frame flicker problem occurs, a new video defogging algorithm is proposed. First, adaptive estimation method of calculating the average video frame atmospheric light value is proposed based on atmospheric scattering physical model, eliminating flicker due to the atmospheric light frame image different from the estimated value of the adjacent emerged. Then, the traditional guided adaptive filtering is weighted. The improved adaptive filter have a guide edge sharpening outstanding features and filtering effect. Finally, according to the characteristics that the weighted guided filter capable of guiding the input image according to the boot image feature, a transmission update strategy based on the weighted guided filter is proposed for fast video defogging. Experimental results show that the proposed algorithm reduces the running time of the video defogging, eliminating flicker and improve operational efficiency.