All, I have been studying the subpixel rendering code and I could spot a few improvement areas and I would like to hear your opinion about it. In this message I discuss one of them.
There are a few issues people face with subpixel antialiasing that yield artifacts on their setup. The most common culprit is gamma mismatches, but quite a few other less obvious factors are involved, such as geometry which is partially addressed. One factor that is not addressed (or even mentioned in discussions at all) is lossy color format and space transformations in display output pipelines. It has become increasingly common that display adapters may autoconfigure the output to use compressed video streams under a lossy (chroma-subsampled) YUV output color format. This is usually done in the context of a trade-off to achieve higher resolutions, refresh rates or bit depths. As an example, many high-DPI panels only offer UHD at 60 Hz under a chroma-subsampled YUV 4:2:0 stream. As another example, many high-refresh rate panels only offer their maximum rates under a likewise compressed format. As yet another example, higher bit depth may also only be available under a chroma-subsampled format. In general, video throughput may be limited by panels, adapters and even physical channels, and the display output may be automatically configured to use a compressed format. As is, the project implicitly assumes an RGB 4:4:4 uncompressed output video format and performs excellently, but if the output has chroma subsampling then subpixel antialiased rendering can exhibit color fringe artifacts. Indeed, several complaints about fringing may very well be caused by chroma subsampling rather than gamma issues. Exposing a subpixel antialiasing filter that addresses cases where the output format has subsampled chroma is quite a large undertaking but is feasible. Though the simpler solution in this case is to switch to grayscale antialiasing. To my knowledge, no libraries commercial or otherwise currently address this issue. Vítor
