Research On Image Visibility Restoration Algorithm In Foggy Conditions

With the rapid development of artificial intelligence technology,computer vision that uses cameras and computer equipment instead of human vision to complete tracking,recognition,analysis,and processing has become one of the most popular research directions in computer applications,and is gradually applied to face recognition,image retrieval,video content understanding,medical image diagnostics,industrial vision,and driverless.However,outdoor computer vision systems are susceptible to weather conditions.The frequency of fog and haze in China has increased in recent years,especially in autumn and winter.In fog and haze weather conditions,the reflected light of the object is scattered and absorbed by aerosol particles(dust and water droplets)suspended in the air before reaching the camera,resulting in loss of image details,decreased contrast and saturation,which hinders the computer analysis and processing of images.Therefore,improving the visibility of foggy images and enhancing the robustness of the system are very important for computer vision systems.This dissertation analyzes and discusses classic traditional methods and deep learning methods,and proposes improved algorithms for the defects of the algorithms.The research in this dissertation is based on the atmospheric scattering model and has obtained certain research results.The specific research work is as follows:(1)Aiming at the problem that the halo phenomenon occurs in the sudden change of the depth of field and the lack of defogging in the distant area of dark channel prior algorithm,a single image defogging algorithm based on fusion and Gaussian weighted dark channels is proposed.Firstly,according to the characteristics of the morphological gradient of the image,the morphological gradient image and the dark channel image are linearly fused to obtain a fused dark channel.Secondly,an adaptive Gaussian weight parameter is constructed to process the fused dark channel image pixel by pixel to obtain the coarse transmittance,and L1 regularization is used to optimize the transmittance.Finally,the fog-free image is restored through the atmospheric scattering model and the restored atmospheric light values.Simulation experiments show that the algorithm can better recover the details of the image and suppress the halo phenomenon.The objective comparison with several typical image defogging algorithms confirms the feasibility of the algorithm.(2)Aiming at the problem of inaccurate selection of atmospheric light values by the existing algorithms,a single image defogging algorithm for near-atmospheric light area detection and transmittance constraint is proposed.Considering that atmospheric light is scattered by the suspended medium in the atmosphere to form background light,which will cause the visibility of the image to decrease.Firstly,the characteristics of the atmospheric light area are used to reversely map the near atmospheric light area in the image to obtain a more accurate atmospheric light value.Secondly,the transmittance is bounded and the near-light area transmittance is optimized,and then artifacts are removed by Gaussian Blur to obtain a more accurate transmittance estimation.Finally,an atmospheric scattering model is used to restore the foggy image.The experimental comparison results show that the proposed method can effectively recover the color and detail of the degraded image,and objective experimental indicators show that the algorithm can effectively improve the visibility of the image.(3)Aiming at the problem that the traditional prior information-based defogging algorithm fails in some special scenarios,an end-to-end convolutional defogging network based on attention mechanism is proposed.The network is mainly divided into two modules: parameter estimation and image restoration.Firstly,multi-scale convolution is used to extract image feature information,and residual network and skip connection methods are used to improve the utilization rate of shallow network feature information.Secondly,the channel domain attention is used to add weight to the feature image input from the previous network and select useful feature information.Finally,the atmospheric visibility model is combined to achieve image visibility restoration.The experimental results show that the proposed algorithm can effectively improve the visibility of the image and the restoration effect is natural.The objective evaluation index of the synthesized image and the real image also shows the effectiveness of the algorithm.