Film Grain Synthesis for AV1 Video Codec [pdf] (norkin.org)

A recent collaboration between Netflix and Google introduced a novel film grain synthesis algorithm for the AV1 video codec, aimed at preserving the creative intent of motion picture and TV content during encoding. The challenge has traditionally been that film grain, while characteristic of analog films, can significantly complicate video compression due to its random nature. The proposed solution leverages an autoregressive (AR) model to transmit grain parameters relative to a denoised signal, optimizing the encoding process and enabling up to a 50% reduction in bitrate for videos featuring heavy film grain. This advancement is particularly significant for the AI/ML community as it enhances the efficiency of video encoding, a critical aspect for streaming and digital content delivery. By integrating flexible modeling of film grain with real-time synthesis at the decoder, the AV1 codec can better accommodate varying noise characteristics and improve video quality without inflating file sizes. The algorithm employs sophisticated techniques like Canny edge detection for film grain parameter estimation and a piece-wise linear function to dynamically adjust grain strength based on signal intensity, ensuring a more accurate and aesthetically pleasing representation of the original film grain in digital reproductions.