is it better having resolution or bits/pixel ?

[sterlina]

As subject says, is it better to mantain a great resolution of the video or lower it to have more bits/pixel?

For better explanation: the bitrate is fixed, so movie size is also fixed.[list=1][*]I got 704*368 pixel each frame, that means 259072 pixels and about 0.257 bits/pixel.[*]I choose to have 640*336 (215040 pixel, that's 83%) and 0.31 bit/pixel[*]even lower resolution: 576*304 (175104 pixel, 68%) and .381 bit/pixel[/list=1]As resolution lowers, I got less details for each frame. But I also got more bit/frame so the picture is less compressed and that means I maintain more details. So... what should be better?

[mpucoder]

Keep the resolution. The way mpeg works, if the picture is too complex for the bitrate then DCT coefficients representing the higher frequencies (ie detail) are removed. This has the effect of lowering the resolution, but only when absolutely necessary.

[Mug Funky]

what bitrate are you shooting for? 1500 kbps? (or do my maths suck?)

that should be plenty for 704x368 using mpeg-4.

[fccHandler]

Here is an observation I made recently:

Traditionally my favorite target for VHS captures is 4700 kbps MPEG-2 video at 720 x 480, plus 384 kbps audio, which will exactly fit two hours on a single layer DVD+R. But recently I had a thought that if I cut the video resolution and bitrate in half, I could fit four hours per DVD.

So, using CCE I did two encodings of the same VHS, one at 720 x 480, 4700k + 384k, and another at 352 x 480, 2350k + 192k. Disregarding the lower resolution and focusing strictly on the MPEG compression, you would think that the 352 x 480 version would look just as good as the 720 x 480 version. In theory the compression should actually perform better, because 352 is slightly less than half of 720.

But in fact, the 352 x 480 version looks visibly worse to me. There is some blocking and compression artifacts which I don't see in the 720 x 480 version! I don't really understand why that is, because mathematically the 352 x 480 version has a higher "bits per pixel" ratio...

But this is what I observed. I recommend you do your own testing and see what you think.

[sterlina]

Originally posted by Mug Funky
what bitrate are you shooting for? 1500 kbps? (or do my maths suck?)

that should be plenty for 704x368 using mpeg-4.

yes, I want to put about 6 hours of video into a DVD, that means a total bitrate of about 1700 (video+audio)

fccHandler: I think the 352 x 480 looks worse because you lower resolution *and* bitrate. So you got (about) the same bit/pixel ratio, but that way you loose half details. The question is: would it look worse if the bitrate was the same?
but you're right: doing test will be a solution.So I'm now doing two encoding, a 704*368 and a 576*304. The first one is done while I'm writing, but have to whait for second one to post results.

A last notice (an obvious one): encoding is much faster when resolution is lowered

[fccHandler]

Yes, some detail is lost because of the lower resolution, but I can live with that. My real complaint was about the MPEG compression artifacts. In both cases the same amount of picture data is given to the compressor to encode at the same bitrate, that is 2350k for 352 x 480, and 4700k for 720 x 480. So if the 720 x 480 version looks clean at 4700k, so should the 352 x 480 version at 2350k. There shouldn't be additional MPEG compression artifacts (blocking and such) in the 352 x 480 version, which is what I observed.

Perhaps the blocking is present in the 720 x 480 version, but the higher resolution is somehow hiding it so I don't notice it as much? That's my best guess...

[mpucoder]

Possibly you notice the blocking more because the picture is resized on playback, and individual pixels as well as the blocks themselves, are larger.

[slavickas]

i wouldn't expect size and bitrate scale at same "speed", testing at same quantizer would be more logical imho

edit: i'm refering to

So, using CCE I did two encodings of the same VHS, one at 720 x 480, 4700k + 384k, and another at 352 x 480, 2350k + 192k

[cyberVera]

For me it is evident that bigger resolution looks better, notwithstanding whatever less of what it gets. It must be the problem of player's resizing in fullscreen mode. But I use 512 or 640 px width, because I noticed that it is easier for computers below 1GHz to playback it smoothly. Quite a strong arguement to count on if you share your movies with others.
This thread is very interesting, I suppose, since it covers the problem without mentioning "compressibility check", which I consider being useless :-)

[sterlina]

Originally posted by cyberVera
This thread is very interesting, I suppose, since it covers the problem without mentioning "compressibility check", which I consider being useless :-) [/B]

well, as I know "compressibility check" is related to the source, and it change as the source is whit more action or slow movement, so you have to run it each time. pixel/resolution tells nothing about a specific encoding but is more general.

I've finally finish my test (my computer was shouted down twice while encoding, one by my brother and one by my father, they didn't see virtualdub ). BUT: there is something wrong because the 576*304 encode is very red colored... maybe related with power off? And their size also differe (about 7%)... maybe human errors? I will run it again, results tomorrow evening.

[Doobie]

To many bits per pixel uses bits inefficiently and wastes them. To few bits per pixel creates a bad picture at any resolution. I prefer to error on the too-few-bits per pixel, given bits-per-pixel vs. resolution.

[Backflip]

Hhmm, what is a good bits/pixel to keep in mind? Some say 2.5 bits/pixel for DivX or XviD, while 2.25 bits/pixel is good for RV9. Something between those targets seem right?

[movmasty]

Originally posted by fccHandler
Here is an observation I made recently:
Traditionally my favorite target for VHS captures is 4700 kbps MPEG-2 video at 720 x 480, plus 384 kbps audio, which will exactly fit two hours on a single layer DVD+R. But recently I had a thought that if I cut the video resolution and bitrate in half, I could fit four hours per DVD.

So, using CCE I did two encodings of the same VHS, one at 720 x 480, 4700k + 384k, and another at 352 x 480, 2350k + 192k. Disregarding the lower resolution and focusing strictly on the MPEG compression, you would think that the 352 x 480 version would look just as good as the 720 x 480 version. In theory the compression should actually perform better, because 352 is slightly less than half of 720.

But in fact, the 352 x 480 version looks visibly worse to me. There is some blocking and compression artifacts which I don't see in the 720 x 480 version! I don't really understand why that is, because mathematically the 352 x 480 version has a higher "bits per pixel" ratio...

EH, you missed another test, that is to CROP the video at 352x480, quality would be certainly better.

Resolution was lower, right, but all the details were still there, just shrinked, so in fact you were encoding the SAME picture to just half the bitrate....
(and,dont forget, vhs is 320x480, so in the 720x480 there was EXACTLY the same amount of details than in the 352x480,
but for a DVD source would be similar if not equal, remember that compression is about ignoring redundant dates...)

Try to zip a bitmap, then the same bitmap with double dimensions, that is 4x the surface, the zip wont surely be 4 times the size of the 1st, because to compress 2x2 pixels of the same colour take the same size than 4x4,
compress a bitmap of uniform color, lets say 4000x4000, 48mB, the zipped size will be close to...nothing.

This is the thing with resolution, compressing a movie with 4x the surface needs only about 2x the bitrate to have the same quality of apparence.

So i like big resolutions at half the frame rate,
and i like to crop, because in this way you really discard details instead to simply shrink them.

But i saw some movie that at 384x288 looked better than one of 640x480 in full screen.

[Wolfman]

I would have thought that video compression is more efficient at the higher bitrate/higher resolution eg there is more scope for compression to happen. ALso keeping the same detail at the lower bitrate will introduce artifacts. Try encoding 4 sets HR/HB HR/LB LR/HB LR/LB (r=res, B=bitrate) and see which is better. I think for divx something between 0.2 and 0.3 q/f (GspoT) gives good quality, generally. How good the motion estimation is also affects this.