Research On Video Image Rain Removal Based On Temporal And Spatial Context Information

With the rapid development of computer technology,image processing and computer vision are widely used in target detection and recognition,intelligent transportation,industrial production and other fields.Clear and reliable image data is particularly important for the development and application of computer vision algorithms.When the image surveillance equipment performs outdoor operations on a rainy day,the captured video image will contain a large number of rain stripes.In the area covered by such rain stripes,the background details of the image will be partially or completely lost,which is not conducive to subsequent image processing.The progress of the work has severely restricted the application of outdoor computer vision systems and computer vision algorithms.This dissertation analyzes the limitations of existing methods,and on the basis of these research methods,applies deep neural networks to the methods of this dissertation.By enhancing the spatial and temporal correlation of information in video images and using visual attention mechanisms,a kind of the video image rain removal method based on the spatiotemporal context information can remove the image rain lines accurately and efficiently while repairing the background information of the rain line occlusion area to a great extent.In this dissertation,we first obtain the prior knowledge and relevant parameters of the rain pattern by analyzing the characteristics of the rain pattern.On this basis,a large data set containing 180 rainy day videos is produced to solve the problems with missing rain / no rain data sets in the deep learning rain removal method.At the same time,this dissertation proposes a method of cascading two four-way(up,down,left,and right)IRNN networks to detect rain patterns in the frame and generate a rain pattern attention map,which can guide the learning of rain / no rain image mapping relationship and remove image rain patterns.And making full use of the correlation of background texture between frames,Three-Round-IRNN background texture repair method is proposed.This method can extract the effective texture features of each frame image for reconstructing the background of the current frame image,and achieve accurate restoration of the background texture of the rain pattern occlusion area..