Research On Rain Removal In Video Image Of Dynamic Weather

Computer vision systems are widely used in intelligent transportation,industrial production,target recognition and other fields.Due to effect of outdoor rain and snow weather,the quality of images captured by computer vision systems is degraded.In addition,rain and snow can cause inaccurate motion estimation of video surveillance systems,reducing compression ratio,restricting the storage system of surveillance video.Therefore,this paper analyzes from the perspective of temporal domain and frequency domain,two different video image rain removal methods are proposed.In order to obtain clear image,the main research contents of thesis as follows:(1)From the perspective of time domain,a video image rain removal algorithm based on raindrops characteristics is proposed.Since raindrops mainly exist in the high frequency components of image,Firstly,this paper analyzes the temporal intensity waveform properties and chromatic constraint properties of raindrops.The method establishes a raindrop model by the temporal intensity waveform properties,the improved k-means clustering method is used to realize the in initial classification of raindrops and background pixels,combining the chromatic constraint properties of raindrops to optimize the results of preliminary classification.Finally,the statistical characteristic is used to calculate the intensity value of the pixels contaminated by raindrops,thereby completing the raindrops removal in the video image.(2)From the perspective of frequency domain,a video image rain removal algorithm based on dual-tree complex wavelet fusion is proposed.Firstly,the image is decomposed into low-frequency sub-images and high-frequency sub-images by the dual-tree complex wavelet.For high frequency sub-images,a fusion rule based on local energy and matching degree is proposed.For low frequency sub-images,a fusion rule based on principal component analysis is proposed.Finally,the image is reconstructed by dual-tree complex wavelet inverse transformation.In addition,in the image enhancement part,this thesis proposes an enhancement algorithm based on bilateral filtering and high-frequency lifting filtering.The experiments results show that the proposed two algorithms have better rain removal effects and robustness.