What is the real visible resolution of a Blu-ray video? (800 * 700?)

[jagabo]

Subtraction shows only the differences but I would prefer to just Interleave() the two videos and compare them visually.

[hello_hello]

I gave it a spin but I still can't get my head around whether it's an effective way to determine resolution in respect to picture detail. Especially as when I tried a section where there were opening credits, the credits appeared to have more picture detail than the picture itself.









To me it's showing differences in brightness more than resolution. And even then I'm not completely sure if that's what I'm seeing. Am I looking at this the right way?

Edit: Damn! I think for the second pair of pics I was off by a frame so they're not the same, but they're probably only off by one so it doesn't change anything in respect to picture detail.

[jagabo]

Titles are often computer generated and overlaid in post. So they are often sharper than the underlying video.

[newpball]

Some people seem to make the mistake of equating video resolution with the actual resolution of a video.

The actual resolution is very often way lower, compressed, anti-aliased and dithered to the max to get some mushy mash that should be taken for high definition.

[hello_hello]

Titles are often computer generated and overlaid in post. So they are often sharper than the underlying video.

So is a measurement of sharpness, if that's what I'm looking at, the same thing as a measurement of picture detail? I'd have thought they're not necessarily the same. Although I'm not sure "sharpness" explains being able to see "detail" around the guy's hand in the first pic (on the left). Neither does detail though. His hand seems neither sharp or particularly detailed. Blurred a bit maybe..... I'm just looking using a CRT monitor at the moment so maybe I'm missing something. I'll look again using the TV tomorrow.

[hello_hello]

Some people seem to make the mistake of equating video resolution with the actual resolution of a video.

Seriously! Coming from the person who's interrupted thread after thread with generalisations about destroying picture detail if you have the audacity to downscale at all, while ignoring examples showing otherwise? Wow! Stick with whatever combination of medication you're on at the moment, it seems to be pretty effective.

[jagabo]

Titles are often computer generated and overlaid in post. So they are often sharper than the underlying video.

So is a measurement of sharpness, if that's what I'm looking at, the same thing as a measurement of picture detail? I'd have thought they're not necessarily the same.

You're right, they're not necessarily the same. Consider a frame that's half white on the left and half black on the right with a sharp demarcation between the two. That video is sharp but has little detail.

Keep in mind that this downsize/upsize technique is not perfect. First of all, any time you resize a video you cause some damage (except for some unrealistic special cases like using a point resize to downscale a video that was upscaled with a point resize). And different scaling algorithms deliver different results. So you will never definitively be able to say that your 1920x1080 video was originally exactly 1280x720. But when downscaling to something larger than 1280x720 the damage won't be too bad. Downscaling to smaller than 1280x720 will show significantly more damage -- assuming that original 1280x720 was sharp and detailed.

In case you're not aware of this: Subtract() in AviSynth subtracts the values of corresponding pixels of the two images. Since pixels can't have negative values 128 is added to the result.

p' = p1 - p2 + 128

So when p2 is bigger than p1 the result is less than 128. When p2 is smaller than p1 the result is greater than 128. And obviously, when they are the same the result is 128. So differences between the two videos show up as variation from a flat medium grey. You can perform a contrast stretch to accentuate the differences.

[hello_hello]

In case you're not aware of this: Subtract() in AviSynth subtracts the values of corresponding pixels of the two images. Since pixels can't have negative values 128 is added to the result.

p' = p1 - p2 + 128

So when p2 is bigger than p1 the result is less than 128. When p2 is smaller than p1 the result is greater than 128. And obviously, when they are the same the result is 128. So differences between the two videos show up as variation from a flat medium grey. You can perform a contrast stretch to accentuate the differences.

I was in fact completely unaware, but now I'm not. Thanks for the info.
Anyway, after newpall's last post I think I need a bit of a lie-down. I'm not coping with the shock as well as I'd hoped.

[hello_hello]

I tried again, and at the moment I'm pretty sold on the "downscale, subtracting the original" method for determining picture detail not being a particularly good one. Not unless there's a resizing method that makes it work. I'll confess I cheated a little with this one as it started as 720p and I resized it to 1080p so I could apply jagabo's script without changing it, but the picture is a little noisy. It's probably digital but in this case there's a scene with enough fine noise that I can see it on my TV from three or four feet away.

Couldn't noise, especially if it's fine noise, be looked at as the ultimate in fine picture detail, even if it's unwanted picture detail, so if the "downscale, subtracting the original" method of determining picture detail worked, shouldn't fine noise reveal itself pretty early? It didn't here. I'm down to 768x320 and there's really not much noise showing, yet in the original video the pattern of noise on the wall can easily be seen.

Am I not looking at this the correct way?

[jagabo]

I see quite a lot of differences in your subtracted image. Noise is low amplitude so it won't show up really well after a subtraction. You can increase the contrast to make the differences more visible. Applying a sharpen filter will increase small detail and noise. Using a sharper resizing filter will also enhance noise and fine detail. Use ColorYUV(analyze=true) to show min/max Y numerically (you'll have to remove the subtitle). Instead of subtracting use interleave and compare the two images directly.

[hello_hello]

Yeah I think I'll go back to comparing with my eyes.

There's lots of differences in the subtracted image, but being a form of fine detail, I sort of thought noise should be one of the first things you would see. The fact that it's one of the last things to show up seems to indicate the subtraction method doesn't really tell you much about the amount of picture detail. Or maybe my theory there is a load of bullocks?

[jagabo]

Another thing you can try to accentuate low amplitude differences is Histogram("luma").

[newpball]

I'll confess I cheated a little with this one as it started as 720p and I resized it to 1080p

Then what would be the point?

For crying out loud, 1080 is twice the resolution of 720, that is significant.

The maximum amount of detail in a video is limited by resolution, but that is the limit, it generally is much worse. Upsampled pictures do not provide more detail, neither does anti-aliasing or dithering or sharpening. In fact they generally destroy detail.

And no, noise is not detail either.

It's obvious to me we have a clear case of not seeing the forest for the trees.

[jagabo]

I'll confess I cheated a little with this one as it started as 720p and I resized it to 1080p

Then what would be the point?

To have a know test cast.

[newpball]

I'll confess I cheated a little with this one as it started as 720p and I resized it to 1080p

Then what would be the point?

To have a know test cast.

I see, so if you have a 720p video, you upscale it to 1080p, then do your test and then you have a good idea of the amount of detail in the video.

What is not adding up to that picture?

[Brad]

There's lots of differences in the subtracted image, but being a form of fine detail, I sort of thought noise should be one of the first things you would see.

The noise in your image doesn't look fine at all to me. It seems to have been hit hard by the compression, turning it into smudgy blobs.

Here are some nice small diagonal lines for Stears555 from a real, official Blu-ray.



Using a sharper resizer makes a big difference, but the finest lines are gone either way.

[Stears555]

Calm down. You were unable to present a test pattern from a blu-ray disk.... After that, you presented a low quality (600 lw/ph) test pattern, which were used in the early stage of HD technology (from the era of HD-DV tapes ). Use a perfect real HD test pattern, with 2000 lw/ph bars and curves and rings....


I believe only for my eyes. The only high quality (200lw/ph) test pattern video -whic was posted by me- shows that the resolution of a blu-ray video is only around 800*600 lw/ph....

[jagabo]

I believe only for my eyes. The only high quality (200lw/ph) test pattern video -whic was posted by me- shows that the resolution of a blu-ray video is only around 800*600 lw/ph....

Maybe you should clarify exactly how you came up with those numbers from your test pattern.

[Brad]

the resolution of a blu-ray video is only around 800*600 lw/ph....

So 480000 lw/ph then?

Show us how you can fit this Die Hard screenshot into 900x700, or 800x700, or 800x600, or whatever smaller dimensions you decide on tomorrow, without losing detail.

[hello_hello]

There's lots of differences in the subtracted image, but being a form of fine detail, I sort of thought noise should be one of the first things you would see.

The noise in your image doesn't look fine at all to me. It seems to have been hit hard by the compression, turning it into smudgy blobs.

You could be right. I was just playing with a video that'd already been resized down to 720p as I didn't have any noisy 1080p handy to play with. It was re-encoded at a very high bitrate though, and the noise does still move quite quickly, which makes it look much finer than a single screenshot might indicate.

I'll try again with some noisy 1080p in the near future though, and experiment with different resizing. Hopefully the result will be different.

[newpball]

It was re-encoded at a very high bitrate though,

May I inquire as to what you would consider a very high bitrate for re-encoding?

[hello_hello]

It was re-encoded at a very high bitrate though,

May I inquire as to what you would consider a very high bitrate for re-encoding?

No.

[hello_hello]

I'll confess I cheated a little with this one as it started as 720p and I resized it to 1080p

Then what would be the point?

For crying out loud, 1080 is twice the resolution of 720, that is significant.

The maximum amount of detail in a video is limited by resolution, but that is the limit, it generally is much worse. Upsampled pictures do not provide more detail, neither does anti-aliasing or dithering or sharpening. In fact they generally destroy detail.

And no, noise is not detail either.

It's obvious to me we have a clear case of not seeing the forest for the trees.

Well you're correct with your last statement but only in that it applies to you.

The object of the excersise was to see if the subtraction method is a good way of determining picture detail. I'd already tried a 1080p source but it was fairly clean. Did you miss that in your rush to share your wisdom? If noise can be considered fine picture detail then the subtraction method should reveal the noise along with other fine detail before anything else starts to show. It doesn't matter if the source video is 400000p 1080p or 480p, the idea is to test how much resolution it has in respect to picture detail.

How does the subtraction method discriminate between actual picture detail and noise? Could you share your wisdom there given you've claimed noise isn't picture detail, which is obvious from a human visual perspective but I'd like to know how a subtraction test knows to ignore it.

The only reason I upscaled to 1080p first is because jagabo's script produces an error along the lines of the resolution not being high enough for the resizing method, or something similar, so rather than fiddle with his script I gave it a 1080p video to start with. I fully expected the subtraction method to go lower than 720p before detail started showing. If you look at the subtraction screenshot you'll see it's down to 768x320 and there's still not much noise showing, or fine detail, although the picture itself it well on the way to becoming defined.

It was a simple test for how the noise would appear as the subtraction method was applied and nothing more. I mentioned upscaling to 1080p first in case there was a reason that'd completely change the outcome in a way I wasn't foreseeing, not to give you an opportunity to state the obvious.

Does that help to clear it up for you?

[jagabo]

The only reason I upscaled to 1080p first is because jagabo's script produces an error along the lines of the resolution not being high enough for the resizing method

At 1 percent the size of the frame drops to zero. Change the script to limit the low values can go. Different algorithms have different limit's but 16 is probably safe:

function DownUpScale(clip v, int percent)
{
  width = (v.width * percent / 200) * 2
  width = (width < 16) ? 16 : width

  height = (v.height * percent / 200) * 2
  height = (height < 16) ? 16 : height

  v.Spline36Resize(width, height).Spline36Resize(v.width, v.height)
  Subtitle("width="+String(width)+" height="+String(height))
}

[hello_hello]

Your second script worked fine with a 720p video, so no need to to resize it to 1080p first. I used the same script while changing the resizing from spline36 to Bicubic and the difference is huge. Spline36 resizes down to about half the resolution before the same amount of detail comes through, and at the same resolution there's a lot less detail.

Now I'm actually using my eyes as well, I'm a little more sold on the idea, but I will play around with it some more. So far, with Spline36 in your script, it seems to match what I'm seeing in respect to picture detail. When I find a resolution where detail just starts to show through, and I then resize the video to that resolution and back again, running fullscreen on my TV I'm struggling to pick the difference between the downscaled/upscaled version and the original (and I'm using Bilinear for upscaling to 1080p with MPC-HC).

When using Bicubic resizing the detail seems to start showing through at a higher resolution, or at the same resolution as spline36 a lot more detail is showing, so I guess using Bicubic instead of Spline36 would be somewhat "safer" in respect to trusting the subtraction test. I'll keep playing.....

PS For the record the 1280x536 video I was playing with was resizing down to 1048x438 without any noticeable loss of detail.

[jagabo]

Yes, different resizers will give different amounts of damage.

Note that I came up with that script off the top of my head -- guessing at Habanero technique. Maybe he'll post back with exactly what he does.

[-Habanero-]

jagabo is correct, that's what my function more or less looks like except I test horizontal and vertical separately and I denoise my footage beforehand so grain doesn't interfere.

Yes, this is a subjective technique and you have to ignore false positives like noise and text. Yes, sharpness is not the same as detail. There's no objective way to measure detail. I just inquired if there was a better technique because I thought the OP's test kid was something new but it just appears to be a test image that the Blu-ray authors say have been mastered the same way as the film on the same disk.