Research On Image De-raining Enhancement Algorithm Based On Deep Learning

The presence of rain water will change the content and color of the image,causing problems such as blurry imagery and dark image color,which will seriously affect human subjective visual perception and outdoor computer vision system tasks.In target detection,rainwater occludes the target in the image,causing the accuracy of target recognition and positioning is reduced.Therefore,effectively remove rain streaks in the image and improve the visibility of the image,which has important significance and practical application value.Existing image rain removal methods have problems such as residual rain streaks in the rain removal results,loss of detailed information,and dark images.In response to these problems,this paper develops a deep learning-based image de-rain enhancement algorithm research.Aiming at the problem of inconsistent rain streaks,which results in poor rain removal effect,the image rain removal algorithm based on multi-stream expansion residual dense network is studied.First,the guided filter is used to decompose the rain image into the detail layer containing the rain streak information.And the base layer containing background information,by directly learning the residuals between the rain and non-rain image detail layers,it can reduce the mapping range and facilitate the low illumination enhancement network to process the darkened base layer.Then use three-channel dilated convolution with different expansion factors to form a multi-scale feature extraction network to extract multi-scale rain streak information in the detail layer;and use dimensionality reduction convolution and dilation convolution to improve the densely connected network and reduce the amount of network calculations,Strengthen the transfer of features,remove rain streaks in the detail layer.In addition,in view of the problem of low visibility in rainy weather,which leads to darkening of the image,the image de-raining network and the low-illuminance image enhancement network are combined,and the image de-raining enhancement algorithm based on the multi-stream expansion residual dense network is studied.Illumination image enhancement network(Retinex-Net)brightens the base layer to improve the visibility of the image background layer.Finally,the detail layer after rain removal and the enhanced base layer are superimposed to reconstruct the result image that both removes rain and brightens.Experimental results on synthetic data sets and real pictures show that the algorithm can effectively remove rain streaks of different densities,improve picture visibility,and illuminate dark details.The comparison with other algorithms also proves that the algorithm has made improvements in subjective effects and objective indicators.