Fog Image Enhancement Algorithm Based On NSCT Transformand Single-scale Retinex

Computer vision has become an important branch of the intelligent field and the basic data source of various disciplines.In the fields of video surveillance,driverless,and threedimensional reconstruction,computer vision plays an important role.However,in actual situations,outdoor imaging equipment is often affected by bad weather,resulting in blurred images that are not consistent with human visual perception.Especially in foggy or hazy days,due to the scattering absorption of light by suspended particles suspended in the atmosphere,the acquired image detail information is reduced and the color is degraded.In particular,the increasingly smoggy weather in autumn and winter in recent years in China has greatly affected the quality of images acquired by outdoor image acquisition equipment.These images obtained in foggy weather often have low contrast,gray color and high brightness.Therefore,in order to make the image acquired by the imaging device in the foggy weather conform to the human eye vision,the image defogging method is required to make the foggy image clear.Observing the images obtained from the haze weather,it is found that the near and far scene information in the image is different in brightness and the details contained therein are quite different.Therefore,this thesis proposes a clear algorithm for the image with uneven foggy concentration.First,an improved single-scale Retinex algorithm is proposed.In the algorithm,the logarithmic function of the traditional single-scale Retinex algorithm and the Sigmoid function are compared.It is found that the Sigmoid function has relatively superior results for the histogram of the foggy image.Due to the characteristics of the single-scale Retinex algorithm,the improved algorithm has a better enhancement effect on the dark areas of the image.Secondly,the luminance component of the foggy image is subjected to multi-scale geometric analysis,and the low-frequency component containing most of the detailed information of the image and the high-frequency component containing most of the edge contour information of the image are processed separately.Most of the noise in the image exists in the edge contour,so this thesis proposes a fast bilateral filtering algorithm,which makes the image edge-protection and reduce the running time of the algorithm.Then,the low-frequency information of the luminance component is enhanced in detail,and a method of balancing the near-field information of the foggy image is proposed.Since the foggy image tends to be too bright,the inverse of the low-frequency component is processed in this thesis,and then the improved single-scale Retinex algorithm with better enhancement effect on the dark region is processed,and the processed image is inverted again,and directly The image processed by the improved single-scale Retinex algorithm is linearly superimposed,and the balance of the far-near view of the enhanced foggy image is adjusted by adjusting the coefficients of the linear superposition formula.Finally,since the color of the foggy image is grayish,a color stretching method is proposed for the saturation component of the foggy image,and the originally smaller color range is mapped to a larger interval.This allows the enhanced foggy image to not only show clear details of the scene,but also to match the visual perception of the human eye.Through the subjective evaluation index,it is found that the foggy image processed by the algorithm has better visual observability and enjoyability.Through objective evaluation indicators,the proposed algorithm has relatively good results in standard deviation,information entropy,peak signal-to-noise ratio and running time compared with other algorithms.