Feeding Longer Frames for Efficient Video Denoising Model

Recently, deep video denoising networks have shown substantially higher denoising performance with considerably lower computing times. However, due to the various characteristics of video motion, such models may not be able to denoise long-term frames. In this paper, we propose a method that takes longer input frames and feeds them to the exist ing architecture. In particular, the proposed method can effectively extract temporal information from frames that are dependent on each other over a long period of time. To demonstrate the performance of the proposed method, we imple mented our method on the state-of-the-art video denoising model. Through the extensive experiments, the proposed method showed better performance in terms of quality metrics than the existing one, even with a higher noise level, resulting in considerably lower computing times.