Research On Image Dehazing Method Based On Deep Residual Learning

Outdoor scene images often experience reduced visibility in the presence of aerosol particles such as smoke,fog,and smog.The reason behind the decline is that the aerosol particles scatter the light reflected from the surface of the object,causing the attenuation of the light intensity.Single image dehazing is an ill-posed problem in computer vision,and its purpose is to recover a clear image from the foggy image.In particular,haze removal can be used as image preprocessing for many advanced vision tasks to improve its visual performance,such as object detection,face recognition,and semantic segmentation.Therefore,single image dehazing has important practical significance for carrying out research on other topics.In recent years,significant progress has been made in image processing methods based on atmospheric physical models.This paper proposes an improved image defogging method based on group sparse representation and an image defogging method based on deep learning.It starts from the principle of foggy image degradation and combines sparse representation and deep learning related technologies.Quantitative and qualitative evaluations were performed to analyze the effectiveness of the method.The main research contents of this article include the following aspects:(1)The single image defogging method ignores the deviation between the natural image and the foggy image,which greatly reduces the accuracy of the coding coefficients during reconstruction and the generalization performance of the defogging model.Aiming at this problem,this paper proposes an improved image defogging method based on group sparse representation.This paper first integrates the degraded model into a group-based sparse representation framework.Then the sparse coefficient residuals of the target image and the reference image are introduced as constraints.Finally,the single image defogging problem is considered as an image restoration problem,and it is optimized using a group adaptive matching algorithm and a group dictionary learning algorithm.This article uses real fog images and synthetic fog images for testing respectively.Experimental results show that the method in this paper has obtained good results in terms of comprehensive restoration of color,details,and structural information.(2)From the physical model of the atmosphere,the influence of haze is inversely proportional to the transmission coefficient of the scene point.Therefore,accurately estimating the transmission map is a key step in reconstructing a fog-free scene.Previous methods used various assumptions or priors to estimate the scene transfer graph.Although many methods have achieved satisfactory results,they are based on strong assumptions and require various parameters related to image formation,which are not always available.For this reason,this paper proposes an image defogging method based on residual network technology,which avoids the estimation of atmospheric light and improves the defogging efficiency.Specifically,first,the network model is designed into two major stages: In the first stage,a foggy image is input,and the network is used to estimate the intermediate transmission map of the foggy image.In the second stage,the foggy image is divided by the output value of the first stage as the input of the network,and the residual network is used to implicitly remove the fog.Finally,the experiments were performed on common data sets,and no-reference index and full-reference index were used to quantitatively evaluate the experimental results.The experimental results prove that the method proposed in this paper has high competition performance.