Haze Scene Restoration Research Based On Model Mapping And Collaborative Network

Image is the basis of vision.For most vision systems,hazy environment will seriously affect the quality of captured images,such as contrast reduction,unclear target subject,loss of details,and low visibility.These systems mostly involve unmanned driving,autonomous navigation,Traffic monitoring,industrial vision,underwater image analysis and many other aspects.In the field of computer vision,in order to ensure the smooth progress of the target task and fully reflect the monitored object,the quality of the image is often required to be higher.Therefore,removing the interference of fog by certain means and obtaining high-quality images is a key preprocessing step in vision tasks.Based on this,through in-depth analysis and research on various existing classical algorithms,this paper proposes a set of new improved algorithms based on traditional thinking and data-driven methods based on the atmospheric scattering model.The specific research results are as follows:(1)Aiming at these problems that the fast median filtering algorithm has poor retention at the edge and many parameters to be estimated,an image dehazing algorithm based on atmospheric veil constraints and piecewise adjustment was proposed.Firstly,basic unequal relation of the atmospheric veil was mapped to plane model of square and its inscribed circle to obtain the constrained initial atmospheric veil.Then the correlation between saturation,depth of field and transmission was used to get a rough atmospheric veil to blur the edge information.Meanwhile,gradient constraint was also introduced to detect the edge cost,which was utilized to correct initial atmospheric veil.In order to improve the distortion of images with close region and the incomplete defogging of images with distant region,an adaptive piecewise adjustment function was proposed to optimize the atmospheric veil.Finally,a clear image was obtained by the restoration module.Experimental results show that the proposed algorithm can be applied to all kinds of images,and has better defogging effect compared to some classic algorithms.The restored image not only retains the information well at the edge,but also the overall color of the image is bright,clear and natural,and the dehazing effect is remarkable.(2)The performance of traditional methods is limited by hand-designed features in the field of dehazing,and problems such as incomplete dehazing and serious loss of details in existing networks,so an end-to-end dehazing model under separated features and collaborative network is proposed.Firstly,traditional atmospheric scattering model is transformed to separate the multiplicative feature and the additive feature.Secondly,according to the influence of two features on final dehazing result,a parallel driven dehazing architecture is designed based on multiplicative and additive feature extraction frameworks.Meanwhile,the spatial information and detailed features of different depths are fully considered in multiplicative feature extraction network,and the purpose of feature reuse and information compensation is achieved by dense cascading to obtain precise and rich target features.Additionally,an additive feature extraction network is built to acquire biased and additive feature according to residual cross-connection structure.Finally,separated features are substituted into restoration model to obtain a haze-free image.Experiments show that the proposed network has significant dehazing effect,natural color of restored image,outstanding detail retention and superior scores of various metrics.