AV1 Repetitive Film Grain Pattern Detection and Quality Assessment

Film grain is a distinctive characteristic of analog film, renowned for its contribution to the cinematic look of videos. Preserving film grain integrity poses a challenge in video distribution, particularly in low-bitrate scenarios, as it greatly restricts compression gains. A V1 film grain synthesis framework addresses this by separating and estimating film grain during encoding, and using these estimates to reconstruct film grain during decoding. The choice of film grain block size, as an important parameter of the framework, significantly impacts reconstruction quality, with larger sizes potentially causing noticeable repetitive patterns, detracting from the viewing experience. In response, we propose a novel first-of-its-kind perceptual quality assessment method to detect and penalize these repetitive patterns efficiently, leveraging the characteristics of the human visual system. Our method normalizes input frames, finds its local similarities with the grain templates to localize repetitive patterns, and uses perceptual pooling to score film grain quality. Implemented in the frequency domain for efficiency, our method was validated through a subjective experiment we conducted with expert viewers. Our experiments show a high correlation between our method and human judgments of film grain quality, providing evidence for the effectiveness of the proposed method.

Published

2024-10-21

Content type

Original Research

Keywords

film grain quality assessment, film grain repetitive pattern detection, auto-regressive film grain synthesis, subjective experiment

DOI

10.5594/MOO/3022

ISBN

978-1-61482-965-2