Research On Single Image Defogging Algorithm

With the development of computer technology and image processing technology,computer vision has been widely used in many fields,such as intelligent transportation,remote sensing,community monitoring,automatic driving etc.However,outdoor computer vision systems are easily disturbed by the external environment in collecting images.Especially in fog and haze weather,outdoor images often degrade seriously,mainly due to the decline in clarity,color distortion and so on.Using hazy images as input will inevitably affect the subsequent image analysis and understanding,and weaken the performance of the visual system.Therefore,it is of great practical significance to study how to effectively reconstruct original images from foggy degraded images to improve the application performance and robustness of vision systems.Essentially,the purpose of image defogging is to remove interference from the bad weather condition,enhance the contrast,recover color saturation,and restore the haze-free image as possible as it can.The main work of this paper is to analyze the imaging mechanism in fog weather and valuable features of hazy images,and seek effective methods to remove haze from foggy images from the perspective of image enhancement and image restoration.The main contents include:(1)To overcome the limitations of traditional image enhancement algorithms in image fogging,a foggy image enhancement algorithm based on scene haze concentration is proposed.In the algorithm,the relationship between the depth distribution of the scene and the image features in three aspects of brightness,gradient and saturation is explored,and the associated multiple feature priors in the hazy image are incorporated to estimate the haze concentration.Based on the mathematical model of total variation optimization,the algorithm separates the texture component from the original hazy image.According to the estimated distribution of haze concentration,the adaptive enhancement on texture component and the correction on color saturation are performed respectively.The experimental results show that the algorithm can restore the characteristics of the scene well,and achieve good subjective effects in improving the clarity and color fidelity.(2)To solve the problem of the dark channel prior invadity in sky area of outdoor images,a foggy image restoration algorithm based on atmospheric scattering model is proposed.The algorithm explores multiple features of hazy images on intensity,connectivity,location as well as probability distribution,and designs a method to recognize sky regions from the input.And the result of sky recognition is used to estimate the global atmospheric light and adjust the lower bound of transmission,which prevents the transmission from underestimation and sky color from distortion.Moreover,in order to increase the computational efficiency of image haze removal,the algorithm processes the hazy image by down sampling to roughly estimate the transmission map and refine the transmission by the minimum filtering and the joint bilateral filtering.The experimental results show that the transmission distribution estimated by the proposed algorithm preserves the scene structure and keeps the local smoothness of the area with uniform depth.The restored images show better visual effect in color fidelity for the sky area and achieve the better real-time performance.(3)To solve the problem that the size parameter of minimum filter template is difficult to be set uniformly,the algorithm based on multiple scales fusion and total variation optimization is proposed to estimate and refine the transmission map.The two transmission maps are roughly estimated from the hazy image by using different template sizes.And in order to preserve the important scene depth edge details and low-frequency smoothing components in the original image,the multi-scale weighted fusion is performed on different layers of Laplacian pyramid decomposition.Besides,the total variation model is adopted to carry out local smoothing optimization on transmission,further eliminating the high frequency texture noise component.And an iterative method based on gradient approximation is also applied to accelerate the solution of the optimization model.In addition,according to the consistency between scene depth distribution and fog concentration,the algorithm uses the multi-prior features of the fog area to estimate the global atmospheric light by locating the deepest pixel position of the scene,so as to reduce the adverse impact of the highlighted object on the estimation of atmospheric light to a certain extent.The experimental results show that the image restored by this algorithm achieves good results in overall contrast enhancement,detail preservation and color saturation restoration.