Image Haze Removal Algorithm Based On Sky Detection And Transmittance Fusion

Images acquired in bad weather,such as haze and fog are seriously degraded,which makes the image color gray,reduces the contrast and makes the object features difficult to identify.That not only affects the human visual effect,but also affects the collection of information in the image,resulting in the computer system being unable to image post processing.Nowadays,bad weather such as fog frequently occurs,and unclear images acquired outdoors increase,and image clarity becomes an important research subject.In order to reduce the influence of foggy images on the image recognition system and ensure the normal operation of the outdoor system,this paper based on the physical model makes an in-depth study on the dark channel prior algorithm and proposes an improved method for the limitations of the dark channel prior algorithm.The specific research content is as follows:(1)Aiming at the problem of misjudging the atmospheric light value in the dark channel prior algorithm,a sky detection method is proposed to obtain the atmospheric light value in the sky region and improve the false value of the atmospheric light value.Firstly,by analyzing the necessity of sky detection and using the Otsu method for sky segmentation,in order to make up for the deficiency of the Otsu method in segmentation,the weighted grayscale method is used to enhance the details of the foggy image,and combined with the value adjustment strategy,morphological erosion and Gaussian blur,the coarse segmentation of the sky and non-sky regions.Secondly,by analyzing the edge-preserving characteristics of fast-steering filtering,the segmentation results of the sky region are finely segmented by using fast steering filtering.Finally,a judgment mechanism is proposed to determine whether there is sky in the image,and the average dark primary color of the sky region is taken as the atmospheric light value.The experimental results show that the proposed method can obtain a more precise sky region,and the estimated atmospheric light value is more accurate than the actual situation,and the algorithm has higher time efficiency.(2)Aiming at the failure of the transmittance of the sky region in the dark channel prior algorithm,a transmission fusion mechanism is proposed to improve the overall effect of image defogging.Firstly,by analyzing the relationship between image brightness information and scene depth,a brightness model is used to replace the sky region depth information,which is applied to solve the sky region transmittance,and the transmittance with luminance information and the dark primary color transmittance are combined.Secondly,by analyzing the characteristics of the transmittance fusion,the source of the fusion weight map is determined,and the fusion weight map is adaptively adjusted by the Sigmoid function.Finally,the fusion transmittance map is refined by the fast steering filter,and then the foggy image degradation model is inverted to perform image defogging.The experimental results show that the image defogging effect of this method is good.The analysis of subjective and objective indicators shows that the method is universal and robust.