Research On Dehazing Algorithm Based On Atmospheric Scattering Model

Computer vision and image processing is an important branch of artificial intelligence,whose main task is to make computers or other artificial intelligence devices perceive the world like humans or other living creatures and have the ability to adapt to the environment autonomously by processing the information captured by image and video capture devices,so as to achieve the goal of computers instead of the human eyes to efficiently complete the identification,analysis and processing of visual information.In recent years,digital image processing has been widely used in target detection,semantic segmentation,scene text recognition and other fields.However,usually,affected by factors such as haze,sand and dust,light sources,suspended aerosol particles such as turbid media in the air greatly hinder the normal transmission of light between the target scene and the imaging device,and the images captured by the visual imaging system are severely degraded,such as image contrast degradation,loss of detail information,scene color deviation,low visibility,etc.,which in turn lead to serious impact on the subsequent work of the computer vision system.Therefore,visibility enhancement of degraded scenes polluted by haze,dust,light sources,etc.has become one of the important research topics in the image processing field in recent years.In this paper,we discuss and summarize the current image defogging algorithms in detail,and propose some improvements and optimizations based on existing technical methods in several aspects,such as transmission estimation and feature information extraction,to address the shortcomings of existing image visibility enhancement methods polluted by haze,light sources,and other factors.The underlying cause of foggy image formation is the scattering and absorption of light between the target scene and the imaging device,and the atmospheric scattering model describes this specific process.Currently,the model is widely used in haze scene restoration tasks.In this paper,two optimization methods for haze,light source image visibility enhancement are implemented based on the atmospheric scattering model,as follows:(1)An adaptive dual transmissions defogging algorithm combined with depth-of-field variation is proposed to address the problems of the existing image defogging algorithm inaccurate estimation of the transmission of the near and distant regions and difficult to adapt to multi-scene defogging.First,by constructing a Gaussian-logarithmic function to map the minimum channel of the fogged image,a rough estimation of the minimum channel of the haze-free image is achieved,while the maximum channel of the haze-free image is obtained by combining the basic inequality relationship of the atmospheric scattering model to realize the dual transmissions;then,the depth-of-field model based on luminance and saturation information is established through feature analysis,and the adaptive transmission function for joint optimization of dual transmissions is constructed by combining the relationship between depth-of-field information and transmission.Finally,the degraded scene is recovered by combining the atmospheric scattering model.The experiments show that the proposed algorithm has clear and natural recovery results,and high visual contrast and rich detail information in the recovery results,and this method solves the problem of inaccurate estimation of transmission in the near and far field areas effectively,and is applicable to multi-scene defogging.(2)To address the problems that the existing image defogging algorithms do not fully consider the image fog information and the nighttime image defogging results are not effective,a novel fog feature map is proposed to reflect the distribution of image fog information,and the unequal relationship constraint method is used to improve the image quality.First,we extract the extreme value channel of the image to achieve a rough estimation of fog information,and optimize it by L-1 regularization to obtain the fog feature map;Secondly,a primary atmospheric light curtain function based on fog characteristics is proposed,and the atmospheric light curtain with fog images after interval constraint is obtained by making an in-depth analysis of color channels and atmospheric light curtain using the mean inequality relation.Finally,the proposed fog information feature map is used to improve the local atmospheric light and realize image defogging based on the atmospheric scattering model.The proposed algorithm is analyzed in comparison with other classical methods on real-world haze images and synthetic dataset haze images,and it is found that the proposed algorithm shows better performance in single image defogging and more advantageous in nighttime haze image recovery.