Research On Video Super Resolution Research Based On Neighborhood Information Learning

High-resolution video provides more detail of the target or scene,and these details play an integral role in areas such as security,medicine,and remote sensing which involving identification,understanding,and analysis.Therefore,the in-depth study of video superresolution reconstruction technology has important theoretical research significance and practical application value.This paper focuses on the research of video super-resolution reconstruction based on sparse representation,and proposes a learning dictionary training method that introduces neighborhood prior information,and applies the learning dictionary to video super-resolution reconstruction to effectively improve the quality of reconstruction;furthermore,aiming at the influence of noise on video,a new approach of introducing low rank matrix decomposition and fusion neighborhood prior information is suggested to perform well in the resolution of video frames.The work of this paper is mainly reflected in the following three aspects:(1)Learning and studying several classical super-resolution reconstruction methods,including wavelet transform method,interpolation method method based on sparse representation,and numerical analysis and experimental results.(2)This paper proposes a training set construction method that introduces neighborhood prior information.The method first extracts the feature points of the video frames to be super-divided,and extracts the block centered by the feature points;secondly,we filter the corresponding points to be over-subscribed in the neighborhood frames.A structural similar block near the feature point of the video frame;finally,a structural similar image block centered on the feature point in the super-subscribed video frame and the neighborhood frame is composed of a training sample set.Ba sed on the training sample set composed of the structural similar image blocks described above,the learning dictionary incorporating the neighborhood prior information is trained and applied to the super-resolution reconstruction of the video.The experimental results show that the introduction of neighborhood prior information is an effective way to improve the quality of video superresolution reconstruction.The learning dictionary trained by this method has low algorithm complexity in the training process,and the dictionary is applied to video super-resolution reconstruction to effectively improve the resolution of the video.(3)In order to overcome the influence of redundant information such as noise on video quality,the low-rank matrix decomposition model is used to remove noise and redundant information in video frames,and the sparse representation super-resolution method based on neighborhood information is further integrated to improve video resolution.The natural and anthropogenic noise is overcome by this algroithm,with which can make a promation to the quality of vidio super resolution reconstruction.