Research On Image Rain Removal Method Based On Deep Learning

The images captured by outdoor image acquisition devices on rainy days often have degradation issues such as background information being obstructed,contrast decreasing,image blurring,and color distortion.These degradation in image quality not only affects the visual perception of the human eye,but also has a negative impact on the operation of advanced visual algorithms in the future.Therefore,utilizing software algorithms for intelligent restoration of degraded images on rainy days is of great significance.This article conducts research on image rain removal methods based on deep learning technology,and the main content is as follows:(1)A multi-scale feature fusion grid rain removal network(MSGridNet)is proposed to address the problem of incomplete and incorrect rain removal caused by inaccurate positioning of rainwater areas in existing image removal methods.MSGridNet first uses a shallow feature extraction module to extract the shallow features of the input rainwater image,and then uses a multi-scale feature fusion module to obtain a multi-scale representation of the rainwater features.Residual learning and smooth expansion convolution are introduced inside the multi-scale feature fusion module to obtain contextual information of rainwater,and channel attention mechanism is introduced to flexibly adjust the weights of features at different scales during the feature fusion process,Accurately locate and remove rainwater areas.The experimental results show that compared to existing advanced methods,MSGridNet has a significant improvement in rain removal efficiency.(2)A conditional generative adversarial rain removal network guided by frequency priors(FP-CGAN)is proposed to address the issue of most deep learning based rain removal methods that only view the rain removal task as a simple end-to-end mapping and do not utilize the inherent prior information of the image,resulting in unsatisfactory generalization performance in real scenarios.FP-CGAN consists of a generator and a discriminator.The generator first maps a rain containing image to a rain free image.The discriminator then uses the highfrequency and low-frequency of the rain free image as additional prior information to score the generator,guiding it to generate more realistic rain removal results.The experimental results show that the method performs better in removing rain on real rainy images than current mainstream methods,and the running speed is also significantly improved.(3)Based on the efficient image rain removal algorithm FP-CGAN proposed in this article,a visualization system for image rain removal processing has been developed..The system consists of four modules: login,image selection,rain removal,and result saving,which can complete the rain removal processing of degraded images on rainy days and improve image clarity.