Research On Rain Removal In Single Rainy Image

Under rainy weather,the images collected by outdoor computer vision systems are sariously degraded by raindrops,which reduce cotrast and blur images,which seriously affect the normal application of outdoor computer systems.In response to this problem,this thesis has conducted in-depth research on rain removal and enhanccment methods for individual images.The specitic work and irmovations arc as follows:(1)Aiming at the problaem of real-time immage clarity for rainy days,a fast rain removal method for single image based on fast weighted median filtering guide is proposed.Firstly,the rainy day image is filtered by the fast weightey median filter,and the raindrops in the image scene are removed;them,the image processed by fast weighted median filter is used as the guide image,and the original rainy day image is guided and filtered to obtain the no rain image that contains edge and texture retention.Experiments show that the proposed method not only can effectively remove the raindrops in the image,but also can protect the texture and detail of the image,and achieve real-time de-rain on rainy images.(2)A rainy day image enhancement method based on multi-stage filtering is proposed for the problem of heavy rain.First,a Gaussian filter is used to decompose the original rainy image into high-frequency and low-frequency parts.Then,a slmilar operation is perfarmed on the high-frequcncy part to further obtain a high-frequency part.The difference between the original rainy image and the high-frequency image yields a preliminary no-rain image.Finally,the dark channel prior as used to defog the no-rain image.The cxperimental results show that the method still has better visual effects under heavy rainfall conditions.(3)De-rain image against night rainy image,a de-rain method based on deep convolutio is proposed.First,the original image is decomposed into a shading image and a reflection image by a relative smoothness method,a reflection image containing a large amount of background details is retained,and a shading image containing a strong light source is removed;then,a multi-scale Retineix algorithm with color restoration the adapted detail enhancement are used respectively.In addition,in order to accelerate the training model,a priori knowledge of the image is needed,so the second guided filtering is used to decompose the enhanced image into the high frequency part and the low frequency part;finally,the high frequency part is taken as the input value of the depth.A no-rain image is obtained and the no-rain image is merged with the enhanced image to obtain a clear rain-free image.Experiments show that this method can not only improve the contrast of the image but also effectively remove the raindrops in the image.