Grainy encodes - Keeping bitrate down.

[PuzZLeR]

I once more call on this forum's great members for a little help.

I've developed a taste for a grainy texture in my video for quite some time now. I personally believe it gives it that cinematic feel, like a professional film look. Grass looks nice, faces look good, backgrounds come alive, etc. And since I been watching stuff on a much bigger screen lately, it's become almost necessary for a good encode - it also does wonders if there were any artifacts in the source.

I do it by either maintaining as much as possible from the source, or adding more via AviSynth or add effects from my editor's features.

All is well except for the "popcorn effect" - the bitrate shoots up dramatically, as is yielded by quantizers and constant quality encodes. This is the case for DivX, even MPEG-2 encoding with CCE's OPV, and ESPECIALLY (using CRF particularly) with x264 - which needs grain the most IMO.

I'm wondering if this is just a fact of life in encoding - that the extra bitrate is necessary - or that quantizers in general are buggy with this. I'm still hopeful that the signal is misled and unnecessarily over-excited by this and can be calmed down - it's not like it's a high motion scene or the like.

A few questions:

How do I control this or keep it down? Is there something I can add, or re-order, in my script?

Or, is there a particular grain filter that will "sneak under the radar" somehow (or something that isn't too bitrate demanding)?

Any help, feedback, or even discussion, is very much appreciated. Thank you.

[redwudz]

Video noise of any type needs more bitrate to faithfully reproduce. Much the same as video of fast moving water or leaves blowing in the wind. There's a lot more detail there compared to a non-moving, simple scene. My guess that's the 'why' for higher bitrate needs. How to 'fix' or minimize bitrate for that type of video I'll leave to someone else.

[jagabo]

High compression codecs get most of their compression from three main properties of video. Highly simplified, these are:

1) Within small areas of the frame there is usually much less information than the number of pixels within that area might suggest. For example an 8x8 block of pixels might all be the same, or nearly the same, color. So you don't need to describe all 64 pixels individually. Far less information can be used to describe them adequately.

2) Major portions of the picture don't change from frame to frame. So most of the time only the changes from frame to frame are encoded. For example, in a talking head shot with a static background maybe only the speaker's lips are moving. The encoder simply says "this frame is the same as the last one, except change this handful of pixels that make up the lips."

3) Often major portions of the image move from one place to another. For example, in a panning shot the encoder can say "shift the image left by 2 pixels then add this column of new pixels to the right edge."

Random static noise (film grain, over the air broadcast static) eliminates all three of these video properties.

[poisondeathray]

Have your cake and eat it too?

How do I control this or keep it down?

You can't have best of both worlds, but here's a few things that may give you some limited control:

In the avisynth filtering aspect, there are many smoothing/degraining filters, and even some that add noise (somewhat unpredictably)

For the encoding part of the equation (for x264), if you've noticed, that crf encodes have increased in size substantially recently. The main reason is that psy-rdo has made it into the main branch and is set at a strength of 1. You can control the strength of psy-trellis and psy-rdo, which you can modulate to enhance grain or reduce grain to your liking. Depending on the build you are using, there are FGO (fine grain optimization) side-builds off the main branch as well, but in testing over at Doom9, it has largely been replaced by psy-rdo and psy-trellis.