The Research On Single Image Dehazing Algorithm Based On End-to-End Reconstruction

With the continuous advancement of industrialization,the density of fine particles in the atmosphere is getting higher and higher.These particles affect the propagation of reflected light from objects,resulting in haze weather,which has become a common weather phenomenon.The haze not only caused great harm to human health,but also seriously affected the effective operation of the outdoor visual system.Therefore,image dehazing has become an important research content in the field of computer vision and image processing.In the early studies of image dehazing,researchers have achieved remarkable results using atmospheric scattering models,but the effect of dehazing depends heavily on the ability to predict accurate atmospheric light and transmittance.In this paper,based on the characteristics of image dehazing and deep learning,a single image dehazing algorithm is studied.The main contents of this paper are as follows:(1)This paper proposes an iterative residual dehazing network.The network mainly uses the image dehazing unit to dehazing the hazy image multiple times using the iterative idea.In the design of the dehazing unit,the idea of long and short-term memory networks and residuals are introduced to further optimize the model.The long and short-term memory networks are used to connect the calculation units at different stages.In the deep processing of the computing unit,residual block connection is used to preserve the original features of the image and prevent over-fitting of the model.This model does not depend on the atmospheric scattering model,but directly generates haze-free images in an end-to-end manner.Experiments show that the iterative residual network can effectively remove the haze in the image.In the test of the synthetic data set and the real data set,the proposed model is better than the existing methods in subjective and objective evaluation.(2)This paper proposes an multi-scale depth information fusion network.According to the atmospheric physical model,converting hazy images to clean images depends on obtaining accurate transmittance and atmospheric light,and the transmittance is directly related to the depth of the scene,so the depth information of the scene is crucial for image dehazing critical.Based on the U-Net architecture,we propose a dehazing network that integrates depth information.The input of the network model is a hazy image.First,the depth information of the hazy image is obtained,and then the depth information is encoded and decoded.In the process,the hazy image features of different scales are skip connected to the corresponding positions,and the output of the model is a clean image.The method proposed in this paper does not rely on atmospheric physical models,and directly outputs clean images in an end-to-end manner.A large number of experiments have proved that the multi-scale deep information fusion network can effectively remove the haze in the image.Not only is it superior to other methods in the synthetic data set,but also performs well in the test set of real scenes.