JCSE, vol. 16, no. 4, pp.185-193, 2022

DOI: http://dx.doi.org/10.5626/JCSE.2022.16.4.185

Feeding Longer Frames for Efficient Video Denoising Model

Kavita Arjun Bhosale and Sang-hyo Park
School of Computer Science and Engineering, Kyungpook National University, Republic of Korea

Abstract: Recently, deep video denoising networks showed substantially high denoising performance with considerably lower computing times. However, such models may not be able to denoise long-term frames appropriately due to the various characteristics of video motion. In this paper, we propose a method that takes longer input frames and feeds them to the existing architecture. In particular, the proposed method may extract temporal information effectively from neighboring frames to address the long-term frame dependency issue. To demonstrate the performance of the proposed method, we implemented our method on opt of the state-of-the-art video denoising model. Through extensive experiments, the proposed method showed better performance in terms of quality metrics than the existing one, even with higher noise level, showing considerably lower computing times.

Keyword: Video denoising; Motion compensation; deep learning; convolutional neural network; noise reduction; signal processing

Full Paper:   415 Downloads, 841 View