Lightweight Video Denoising Using a Classic Bayesian Backbone
This addresses the trade-off between denoising quality and speed for video processing applications, offering a lightweight solution that is incremental over classic Bayesian methods.
The paper tackles the problem of high computational cost in state-of-the-art video denoising by proposing a hybrid Wiener filter with small ancillary networks, achieving performance within 0.2 dB of a leading transformer method while being over 10 times faster and using fewer parameters.
In recent years, state-of-the-art image and video denoising networks have become increasingly large, requiring millions of trainable parameters to achieve best-in-class performance. Improved denoising quality has come at the cost of denoising speed, where modern transformer networks are far slower to run than smaller denoising networks such as FastDVDnet and classic Bayesian denoisers such as the Wiener filter. In this paper, we implement a hybrid Wiener filter which leverages small ancillary networks to increase the original denoiser performance, while retaining fast denoising speeds. These networks are used to refine the Wiener coring estimate, optimise windowing functions and estimate the unknown noise profile. Using these methods, we outperform several popular denoisers and remain within 0.2 dB, on average, of the popular VRT transformer. Our method was found to be over x10 faster than the transformer method, with a far lower parameter cost.