Tweet
Codecs: Battle of the postprocessors?
I've been studying DivX 5.1 and the upcoming MPEG 4 Part 10/AVC/JVT/H.26L/H.264/whatever-the-****-they-call it-now, and my impressions have been pretty muted. Divx looks better without any postprocessing at low q (31 quant); my brain can decide best which detail is important and which is pure noise. JVT, on the other hand, cheats and seems to include PP in the decoder itself, leading to a smoother, but less defined image. If I had to watch one or the other, I'd go with DivX. But holy hell, JVT has great de-ringing! Solid text looks quite awesome in most circumstances.
At higher quality levels (14 quant), the playing field gets a bit muddier. DivX with PP on is beaten by JVT in frame-by-frame comparisons. Without PP, however, the lack of blending exposes discoloration & blocking, while bringing the edge definition to a *bit* above par. Of course, when played at 30fps, you can't quite appreciate that definition, and the faults become dominant. I can't do a proper analysis of DivX in full motion w/ high PP (slow computer, and Vdub won't respect the PP settings), but it would probably come close to JVT. In short, I think JVT does a better job of hiding compression artifacts (through built-in postprocessing, I assume), at very slight cost of definition in all but the lowest quality encodings, while DivX maintains that extra bit of definition by default, exposing artifacts, and has comparatively lousy smoothing/deringing functionality.
JVT has been a while in the making, so I was hoping for markedly better results, but I'm now starting to suspect that the future of video codecs may lie largely in intelligent postprocessing. This is handy for broadcast use, since the common eye prefers to have some blur instead of evident artifacts--though people looking for details might think otherwise. An intelligent PP will make them both happy, which is why I think it's a grossly underappreciated part of many codecs today. Any thoughts on this?
(BTW, has anyone looked at VP lately? 3.2 sucked in all areas but processor usage.)
At higher quality levels (14 quant), the playing field gets a bit muddier. DivX with PP on is beaten by JVT in frame-by-frame comparisons. Without PP, however, the lack of blending exposes discoloration & blocking, while bringing the edge definition to a *bit* above par. Of course, when played at 30fps, you can't quite appreciate that definition, and the faults become dominant. I can't do a proper analysis of DivX in full motion w/ high PP (slow computer, and Vdub won't respect the PP settings), but it would probably come close to JVT. In short, I think JVT does a better job of hiding compression artifacts (through built-in postprocessing, I assume), at very slight cost of definition in all but the lowest quality encodings, while DivX maintains that extra bit of definition by default, exposing artifacts, and has comparatively lousy smoothing/deringing functionality.
JVT has been a while in the making, so I was hoping for markedly better results, but I'm now starting to suspect that the future of video codecs may lie largely in intelligent postprocessing. This is handy for broadcast use, since the common eye prefers to have some blur instead of evident artifacts--though people looking for details might think otherwise. An intelligent PP will make them both happy, which is why I think it's a grossly underappreciated part of many codecs today. Any thoughts on this?
(BTW, has anyone looked at VP lately? 3.2 sucked in all areas but processor usage.)