JVET-AF0209 AHG13: Frequency domain Film Grain Objective Metrics X. Meng, W. Zhang, S. Labrozzi (Disney Streaming) [late]

In this proposal, a subband based film grain assessment method (SFGA) is proposed, which shows promising results. SFGA, designed as a full-reference film grain quality assessment tool, initially transforms both reference and test images from the spatial domain to the frequency domain. It then selects several sub-bands within the frequency domain, compares their sub-band noise power spectra, and subsequently combines the sub-band results. The PLCC and SROCC between the proposed metric and MOS are both larger than 0.75, outperforming existing metrics like NSS (Natural Scene Statistics), FFT (Fast Fourier Transform), PSNR, and SSIM. This metric not only enhances our understanding of film grain technology, but also provides a valuable tool for fine-tuning film grain parameters in the modelling and synthesis process.

Subbands with 4 orientations and 4 scales.

Availability of objective metric would be highly desirable. This is a very attractive approach.

It was asked whether software is available. The proponent responded that some legal issues would need to be resolved, hopefully by the next meeting. If data are generated by JVET, the analysis could be made by proponents.

Is the analysis done only in luma or in RGB domain? Only luma, even though the noise is coloured.

It was commented that evaluation when compression artifacts are coming in would be desirable.