Research Of Rain Removal Method For Single-image

On rainy days,the captured images and videos are susceptible to be affected by the scattering and fuzzy effect of the raindrops,which makes the image blurred and visibility decreased,therefore,it greatly limits the performance of the outdoor visual processing algorithms,such as image segmentation,target recognition,target tracking and so on.At present,most of studies have been done based on video rain removal.Unlike the video rain removal,the researches of the single image rain removal have aroused the concern of domestic and foreign scholars for lacking of prior information of times and places.This paper focuses on the single image removal whose implement process and innovation can be concluded as follows:Firstly,combined with the advantages of propagated filter and guided filter,a new method is proposed to decompose the rainy images.After being decomposed,the rain streaks in the high frequency image are kept intact and the details of the background are obvious in the low frequency image.Secondly,we present a method for extracting the pure rain region in the high frequency gray image by using the gradient amplitude and gradient direction on the edge of the rain streaks.The learning of rainy dictionary can be achieved based on the extraction of pure rain area.The high frequency gray rainy image can reconstruct by using the learned rainy dictionary.Then,the rainy mask can be obtained through binary process.Before accurate detection of rain streaks can be achieved,the rainy mask is optimized according to the color characteristics of the rain streaks.Finally,single image rain removal can be gained with the method of the improved bilateral filter based on the rainy mask and the characteristics of the rain streaks.This kind of filter method can not only deal with rain pixels and rainless pixels personally,but also can effectively eliminate the influence of different target objects on center pixel filter,so as to improve the visual effect of de-rained high frequency image.A rainy image can be recovered by combining the de-rained high frequency image with low frequency image.The experimental results show that the method proposed in this paper can effectively detect the pure rain area in the rain image,and the rain detection and removal is more accurate.Compared with the existing algorithms,the visual effect of the rain removal image is better.Moreover,our work is not limited by the directions and scales of rain streaks,so it can be applied into more applications.