A Perceptual-based Neural Video Preprocessing Hardware Solution for Video Encoding Applications
Pierre-Alain Afro, Loïc Strus · 2025
In recent years, numerous studies have focused on improving video quality, particularly by optimizing video coding for perceptual enhancements, as reviewed in [1]. Many studies focus on Convolutional Neural Network (CNN) based in-loop filters such as [2] or post-processing filters as in [3]. While these approaches are promising, being outside of the coding loop allows for quicker innovation and deployment, as it can operate with existing installed decoders without requiring modifications to the standards. On the other hand, research on CNN-based pre-processing for video coding has emerged as it does not affect any potential standards. This approach aims to save the most bitrate while maintaining the same level of perceptual quality.