This paper proposes a novel real-time video denoising method, ALTA-PGD, leveraging tensor singular value decomposition to exploit low-rank structure of video data. An adaptive tubal rank selection mechanism and accelerated proximal gradient descent algorithm are introduced for efficient computation and improved convergence rate.
Key findings
Proposes ALTA-PGD for real-time video denoising leveraging tensor singular value decomposition.
Introduces adaptive tubal rank selection based on local noise levels and motion characteristics.
Employs accelerated proximal gradient descent with Nesterov momentum for improved convergence.
Aims to process high-definition video at 30 FPS on modern GPUs with competitive denoising performance.
Limitations & open questions
The paper is a research proposal and does not include experimental results.
The effectiveness of the proposed method in real-world scenarios is yet to be validated.