[MEncoder-users] encoding technical question

[A. Schmid]

Philippe MONROUX wrote:
> De (from) (von) <bugfood-ml at fatooh.org> :
> 
>> Philippe MONROUX wrote:
>>>> 2) Unless I am very much  mistaken, CQ will not give you predictable
>>>> file sizes, since it will  use whatever bitrate is needed to provide
>>>> the quality specified., so the  basis of your question is flawed. In
>>>> my experience, the only way to achieve a fixed final file size is to
>>>> do two pass encoding specifying bitrate, not CQ or CR.
>>> Sure but, read in doc :
>>>
>>> resolution x CQ = (bitrate x 1000)/25 (if 25 fps)
>> Where is this written? 
> 
> Here :
> http://www.mplayerhq.hu/DOCS/HTML/en/menc-feat-dvd-mpeg4.html#menc-feat-dvd-mpeg4-resolution-bitrate-compute

I think there was some confusion with the CQ, which means "compression 
quality" in this document. It is stated clearly that "CQ represents the 
number of bits per pixel and per frame of the encode".
This means CQ ist in fact an way to talk about constant _bitrate_, not 
constant quantizer (or "constant quality") - which is also abbreviated 
"CQ" but means variable bitrate.
So just replace CQ with BPP (bits per pixel) and everything should be fine.

> The ResY formula (without 16 rounding) is
> 
> ResY = SQRT( 1000*Bitrate/(25*ARc*CQ)) or
> 
> (ResY)^2 = 1000*Bitrate/(25*ARc*CQ) or
> 
> CQ*ARc*(ResY)^2 = 1000*Bitrate/25
> 
> And since ResX = ResY * ARc (without 16 rounding too) that give :
> 
> CQ*(ResX/ResY)*(ResY)^2 = 1000*Bitrate/25
> CQ*ResX*ResY = 1000*Bitrate/25

This formula has some truth in it, but I wouldn't take it for THE 
perfect formula that always gives a perfect encoding.

It has been found out, that BPP values between 0.1 and 0.3 are the only 
sane values (for today's codecs).

In http://www.mplayerhq.hu/DOCS/tech/encoding-tips.txt you can read:
< 0.10: don't do it. Please. I beg you!
< 0.15: It will look bad.
< 0.20: You will notice blocks, but it will look ok.
< 0.25: It will look really good.
 > 0.25: It won't really improve visually.
 > 0.30: Don't do that either - try a bigger resolution instead.

You formulas just calculate the resolution, based on the BPP you want 
(quality) and the bitrate you wish to spend (size), while keeping an eye 
on video and pixel aspect ratios.

This is a little problematic, because clearly quality and filesize _must 
be_ conflicting properties! When keeping the original resolution, you 
can only choose one of them (as you somehow do with constant 
quantizer/VBR (choosing quality) opposed to constant bitrate/CBR/ABR 
(choosing size)), the other one then depends on the first one.

The author finds a way out of this dilemma by varying the resolution. Of 
course this works, but a lower resolution also means less detail, very 
much like higher compression. So in the end it boils down to filesize 
versus quality, again.

As for your original question: I would suspect that 600x200 with CQ=0.10 
will look just as bad as 450x150 with CQ=0.177 _on average_ depending on 
the movie characteristics).
The difference is, that the latter will have less artifacts! Both will 
lose about the same amount of detail (the first because of higher 
quantizers, cutting off more of the high frequencies, the second because 
of the sampling theorem, which just "forgets" about the high frequencies 
and leaves more bits for encoding the low ones).
It's just that you don't see the lost frequencies on nicely (= low pass 
filtered) scaled down images, whereas you DO see blocking artifacts on 
images with quantizers too large!

This means the hi-res/low-bpp approach looks better on images where you 
need fine details but little motion (e.g. presentations), 
low-res/high-bpp looks better on high motion scenes where fine details 
aren't important (e.g. action movies).

Regards
Alex