Research On Image Clarification Methods For Different Weather Outdoor Image Based On Learning Idea

From low-level image processing technology to high-level reasoning task,image clarification involves a wide range of research fields.The reflecting factors of weather types such as rain,snow,haze,shadows,etc.are involved in many visual tasks such as disaster warning,automatic driving,scene understanding.These factors will not only cause image quality degradation,but also make image deformed,blurred or even invisible.If the vision system cannot remove the above factors,the above factors will affect the effects of visual tasks such as recognition,tracking,understanding,and even lead to the failure of related vision tasks.Image clarification has become an urgent problem in the field of visual processing field.In view of this,this paper aims to study the theories and methods of removing the factors that affect the clarity of outdoor images in different weather conditions and improving the visual effects of images.The specific research work is as follows.First,in view of the limitation that existing weather classification methods can only classify weather in fixed scenes,the starting point is to study the factors that influence the clarity and visual effects of outdoor images in different weather,and a weather classification method for outdoor images is explored.Based on graphics theory,the imaging characteristics of each factor are analyzed,and a multi-branch and multi-passby feature extraction method is designed.Aiming at the distribution characteristics and internal relations of each factor in the image,a feature weighting scheme based on the attention mechanism is formulated to assign weights to the extracted features.Combining the corresponding semantic information of each factor in the outdoor image,the feature enhancement method is designed to enhance the expressive ability of the extracted features.Combining the designed multi-branch and multi-passby feature extraction method,the feature weighting scheme based on the attention mechanism,and the feature enhancement method,the weather classification model is constructed to distinguish the type of weather in outdoor images,which provids theoretical support for the research of image clarification methods.Secondly,to remove the factor of sunny days,that is,the influence of shadow on the visual effect of outdoor images,and to solve the problem that the existing shadow removal methods cannot effectively deal with the shadow boundary,a shadow removal method combining manual features and neural network learning features is designed.Analyze the imaging principle of shadow,based on knowledge transfer and end-to-end idea,a convolutional neural network is constructed to extract neural network learning features related to shadow,and to obtain the mapping relationship between shadow images and shadow-removed images.According to research the imaging characteristics of shadow boundary pixels and non-shadow boundary pixels,design and define the brightness-gradient difference feature as the manual feature of the shadow removal problem.Aiming at the essence of shadow removal problem,integrate the designed brightness-gradient difference feature with neural network learning features and a loss function is construct which optimizes the mapping relationship between the established shadow image and shadow-removed image,and improve the effectiveness of shadow removal method.Thirdly,to solve the problems of the universality of the existing outdoor image removal methods and the need to improve the clarification effect,a rain and snow removal method for outdoor images is researched.Based on the principle of optical imaging,the distribution characteristics and optical properties of rain and snow in outdoor images are analyzed,and a model that can remove rain and snow in outdoor images is constructed based on the principle of encoding and decoding.The loss function is constructed based on deep learning to optimize the clarification of rain and snow model,which achieves the removal of rain and snow in outdoor images.Based on the ideas of low-pass filtering and mean filtering,an image enhancement scheme is designed to further remove the rain and snow in real-world images,and optimize the effect of removing rain and snow in the real-world image.Then,to solve the problems of insufficient universality of existing outdoor image dehazing methods and poor image visual effects after dehazing,the dehazing method for outdoor image based on the idea of generating adversarial is researched.According to the atmospheric scattering model,the imaging principle of haze in outdoor images is analyzed,and the mapping relationship between haze image and haze-removed image is obtained based on adversarial learning idea.According to graphics theory and fusion of structural similarity ideas,a loss function is constructed to optimize the mapping relationship between haze image and haze-removed image.According to the idea of contrast enhancement,an image enhancement scheme is designed to optimize the dehazing effect for real-world images.Finally,in order to break through the current situation that the existing image clarification methods can only deal with single factor,an image clarification method that can remove four types of factors based on deep learning and image decomposition is designed.According to the optical scattering theory,the internal connection and external difference between rain,snow,and haze is analyzed,and the calculation method of global atmospheric light value and transmission map is designed.Combining deep learning idea,a model for calculating rain and snow layer is constructed,and by fusing the designed global atmospheric light value and transmission map calculation method,effective processing of rain,snow,and haze in outdoor images is achieved.Based on the shadow imaging principle,the correspondence between the RGB channels is established,combine the image orthogonal decomposition idea to solve the established channel correspondence,use the Gaussian mixture model to obtain the shadow area,and remove the shadow in the image based on the mean filtering idea to achieve the processing of rain,snow,haze,and shadow in outdoor images.