Resolution vs Bits/Pixel

[Bree]

Has anyone performed a solid comparison between the trade of resoultion vs Bits/Pixel. What I mean is given the following fixed starting points:

DVD 720x480, progressive 24 fps source
Moderate motion movie

Fixed End Point:
Encoded size is the same

Now vary:
Encode Codec: MPEG1 or MPEG2
Encoded Resoultion
Bit/Rate up to a max of 2.5Mbit/s

For instance and XVCD, MPEG-1 at 352x240 at 2500kbit/s has the same file size as an SVCD, MPEG-2 at 480x480 with 926kbit/s

Likewise a SVCD at 360x480 with 2500bit/s has the same file size as an SVCD at 480x480 with 1875kbit/s

Better to throw a higher data rate at a pixel or more pixels with lower data rate????

[Truman]

Resolution of a MPEG does almost not effect the filesize.
The more pixel (actually: macroblocks) you have to encode the higher bitrate you need to keep the bits per macroblock on an acceptable level.

[Bree]

Truman, I'm confused. Doesn't your first sentence contridict the second. You say resoultion doesn't matter to file size. ie 352x240 res. filesize = 480x480 res. file size, however a 480x480 image has many times more pixels than a 352x240 image. If a higher data rate needed for more pixels than doesn't that equate to a large file. or conversly poor image quality for the same file size?

[Truman]

The resolution itself does not effect the filesize.
For instance and XVCD, MPEG-1 at 352x240 at 1150kbit/s has almost the same file size as an SVCD, MPEG-2 at 480x480 with 1150kbit/s.
At that low bitrate the 480x480 MPEG will look worse, you should increase the bitrate to get at least decent quality, then the filesize will get larger.

[Bree]

Ok, I think I may begining to see.
So then given the following:

type1 352x480 SVCD at 2500kbit/s
type2 480x480 SVCD at 2500kbit/s

1. will the file sizes be approximately the same?
2. which will yield the better quality?

[Jumbalaya]

I think the idea that Truman is trying to get across is that bitrate is the ONLY determination of file size. Think about what bitrate means: the amount of data per unit time. So, if you have the same bitrate and the same playback time, the final size of the file will be the same. When you tell the encoder to use 2500 kbit/s, you are, in fact, restricting the encoder to produce a file that averages about 2500 kbit/s. This means that no matter what resolution you choose, the final file size will be roughly the same.

type1 352x480 SVCD at 2500kbit/s
type2 480x480 SVCD at 2500kbit/s

In your example above, both files will have the same file size. However, type2 could potentially look worse because of more macro-blocking. When you increase the resolution without increasing the bitrate, you're telling the encoder to do more with what it has. A higher resolution requires more space to encode per unit time, but you're not giving it any more room to work with. It'll have to start throwing more stuff out, potentially causing more macro-blocking.

Looking at your original question, it seems that you were using the rule of thumb that "higher resolution = better quality". This is true, but only if you throw enough bitrate at it.

[mickboss]

8) BIG DEAL: If you use 480x480 or 720x480 you will notice that 720x480 will give you A LOT of resolution more (believe me, you´ll notice) and compression artifacts have almost no difference!

The best thing you can do is to try. I already tried and I concluded:

- It´s best to encode 720x480 (in my case 720x576) with movies with low/moderate action (Image quality improves a lot and almost no difference with macroblocks)

- It´s best to encode 480x480 (in my case 480x576) ONLY with movies with A LOT of action


Usually I use 720x576. Only with my homemade VHS/HI8 videos I use 480x576

Try and find out

[KoalaBear]

For instance and XVCD, MPEG-1 at 352x240 at 2500kbit/s has the same file size as an SVCD, MPEG-2 at 480x480 with 926kbit/s
Likewise a SVCD at 360x480 with 2500bit/s has the same file size as an SVCD at 480x480 with 1875kbit/s

I think I see what you're getting at.

Every MPEG resolution has a "transparency bitrate" associated with it, i.e., a bitrate at which the copy is indistinguishable from the original. The DVD spec assumes CCIR-601 is transparent above 9 Mb/S, so using that as a baseline we can estimate the transparency bitrate for other resolutions proportionally:

	Resolution	Size	  Transparency
	----------	-------  ------------
	CCIR-601     720x480  9 Mb/S
	2/3-D1       480x480  6.75 Mb/S
	1/2-D1       352x480  4.5 Mb/S 
	1/4-D1       352x240  2.25 Mb/S

What this means, basically, is that the only platform capable of delivering the majority of these resolutions transparently is DVD itself.

However, there are tricks that can be used to increase the effective bitrate. NTSC-FILM (24fps) is 20% more efficient than NTSC video (30fps); a 352x480 frame size is 22% smaller than 480x480; letterboxing can reduce the effective resolution another 35%, and so forth.

Thus, a 352x480 NTSC-FILM MPEG at 2.3 Mb/S has an effective bitrate of 3.2 Mb/S fullscreen or 4.0 Mb/S letterbox, very close to the transparency bitrate for that resolution yet within the range of what X[S]VCD can deliver.

Now, as to the question of file sizes, this is where I'm a bit confused. The length of an MPEG recording doesn't necessarily reveal its quality. Generally speaking, the higher the bitrate the better, but a well-crafted VBR MPEG can look much better than a CBR MPEG of the same length because the bits are distributed more efficiently.

Am I getting warm here, or have I missed the point?

[Bree]

I think I'm getting it. I do understand intuitively that for a given resolution, in general it is better to through a higher bitrate at the encode (benefits of VBR vs CBR aside). However, how far should that be taken wrt resolution. What I mean to say is bit rate / pixel proportional to quality? I belive your example of transparency would imply that there is. So for discussion sake: Let's stick to encoding SVCDs, and a player that has a max playback capability of 2500kbit/s.
so encode 2 pass VBR and only chane resoltion:
720x480
480x480
352x480
352x240
64x64
?x?
is there a rule of thumb for which will be better and why. Thanks

[Bree]

Ok I did a quick and dirty experiment.
720x480 at 2500kbits
480x480 at 2500kbits
352x480 at 2500kbits
352x240 at 2500kbits

In the extreme 720x480 looked "sharper" than 352x240. By that I mean, edges were well preserved fine detail was present. However in comparison the 720x480 had more "noise" than the 352x240. There were DCT blocks visible in the 720x480 but not in the 352x240.
As far as the middle 2 resolutions, it was fairly hard to distinguish a "noise" difference between the 720x480,480x480, and 352x480 encodes. However a slight degredation in the "sharpness" was visible from 720x480 to 480x480 and hardly any difference between 480x480 and 352x480.

This was a 30sec scene with some moderate character motion but no camera motion.

Overall, qualitatively I would have to say that the 720x480 looked the best. However my concern is that with motion that it may get ugly quick with artifacts.

Basically I would describe a quality factor as a composite of 2 factors:
Resolution and SNR(signal to noise ratio)
Resoultion (sharpness, detail, edges, etc) = size (ie 480x480, etc)
SNR (DCT blocks, encode errors due to motion) = kbits

Now to come up with a mathematical representation of quality such that a good encode can be had every time......

[Kdiddy]

waaaaayyy too much time on your hands...LOL

[droolian01]

I had difficulty at first understanding this concept, until i read this excelent article:-
http://nickyguides.digital-digest.com/quality.htm

It seems that the resolution you choose dictates the number of 16x16 macroblocks there are in each frame (wow - just figured thats why source vid should have a resolution that is divisible by 16!!! - cool thanks!)
and each frame can only have a fixed amount of data allocated to it, depending on the bit rate. Your experiments indicate to me that there must be an optimum resolution - bitrate relationship, probably depending on how much 'motion' there is in the whole film, so universal figures are likely to be impossible to evaluate.

But of course beauty is in the eye of the beholder, and thats what makes these forums interesting!

[droolian01]

Sorry, in the link above read the subsection called 'quality and resolution'

[Bree]

Good article!

they actual posted examples of the opposite test that I did. Which is to say, they kept resoultion constant and varied bit/rate. I could post my results of constant bit/rate and varying resolution if anyone is interested.

If you are going after best quality, and filesize is not an issue, and given the limits of most DVD players for playing SVCDs (ie CD read speed, ie bitrate limitation). Then I really think the more realistic trade is to vary resoultion to improve (optimize) video quality.
As pointed out early by Kolabear that encoding at 24fps and widescreen are easy ways to free up bitrate for a given resolution.