A Video Denosing Technique Employing Advanced Learning and Motion Estimation Method

Images acquired by cameras contain noises for various reasons. Removing noise is essential in computer vision, such as high quality video object recognition. Much researches have been done on noise removal for frame by frame base, but there are little researches on noise removal on video, which employs temporal correlation of the video sequences. Recently, deep learning-based video sequences noise removal has been introduced , but sufficient consideration of video motion has not yet been made. Thus, in this study, we propose an improved noise removal techniques based on advanced motion processing and learning method. The proposed method is based on the recently introduced video noise removal techniques known as DVDnet and fastDVTnet. Using some modified leaning functions, the proposed scheme shows superior performance in video noise removal with respect to conventional methods. Computer simulations based on deep learning with reasonable dataset shows that the proposed method has better noise removal results with respect to conventional methods.