1. Hi, this could be an unusual request.
I come to the point:

- I have the DVD of a movie, a typical PAL DVD SAR 64:45, anamorphic res 1024x576.
- I also have a 1920x1080 WEBDL of the same movie from the stupid national broadcasting network, released on their stupid website.
- I say stupid cause the AR of the WEBDL if totally wrong.
- My objective is to re-encode the WEBDL changing its AR in order to match the DVD one.

I have asked in a local forum, and they suggest to find a frame with a sphere or a circle perpendicular to the video shooting plane,
and try several "resizings" until it appear a perfect circle or close to it.

I've answered: "ok, I got it and I think I could do it".
Anyway, apart the fact that a sphere is not always available, this method looks to me a bit tricky and imprecise.
Then I started thinking and reasoning about this, trying to find a better, more reliable and practical system.

What I though is:
- Sources have different crop.
- I have the DVD then I know the correct AR.
- I have both DVD and WEBDL objects resolutions, so I'm quite sure I can use a mathematical proportion
to calculate the correct correction ratio to apply on the damn WEBDL.

Continuing on this road, I've taken two screenshots of the same frame. (this is only an example, not the definitive frame.)
https://www.dropbox.com/sh/qpcoh9bbwtkd69j/AACCFiFZMHDY7Sf3M-4Rqkawa?dl=0

Basically I've chosen this frame cause I can see light reflection in both characters eyes, that I want to use as reference points,
but this can do with every else object.

Subsequently I've made two almost identical selection with an image editor on each ss, the editor gives me the AR of the selections.

At this point I start calculations:
- 2.096 / 1.892 = 1,1078 (rounded)
- 1920 / 1,1078 = 1733 (rounded) = 1732 (rounded to the closest mod4 res)
- that's it: 1732x1080 is the almost correct new resolution for the WEBDL.

here's the ss resized to 1732x1080:
https://www.dropbox.com/s/n2wpzck3sdgc103/bn%201080%201732%20x%201080.png?dl=0

My questions are:
- this can be considered a good and reliable method, or am I forgetting something?
- I suppose that higher is the distance between the two reference points, much more is the accuracy, right?
- Can you suggest another, more precise, reliable or simpler method in order to achieve the objective?
2. You might have arrived at the correct AR of the WEBDL. You would need to check sources such as IMDB to confirm that 1.6:1 is correct.

However, you are basing your calacluations that the dvd is properly displayed. If it was then that would also be 1.6:1 execpt that it is not.

Likely that the dvd was framed differently to avoid excessive borders. Note there is a very slight one both left and right. Not present on the web version. And equally the web suffers from minor but still annoying stretching.
3. something escapes me..
excuse me DB83, but both the AR of the selected areas (which have eyes as corners) of the DVD and the resized WEBDL are the same,
is about 1.888, is this not enough to say that now the resized WEBDL has the same AR of the DVD?
4. Maybe I did not explain myself well. The dvd is 1.78:1 (standard 16:9). But if the original image was 16:10 or 1.6:1 and slighly zoomed in to avoid borders then both are indeed 16:10. It is the dvd AR that is then fooling you.

BTW What film is this and then we can check the correct AR.

Now I did not check the edges of both initial frames to see if the is indeed info missing on the dvd*

As for the 1.888 not sure where this comes in to the calculations.One reserve I have is that the level of magnification is different.

*edit: Actually there is more information on the dvd. So the images are not identical in the first place
5. Originally Posted by DB83
Maybe I did not explain myself well. The dvd is 1.78:1 (standard 16:9). But if the original image was 16:10 or 1.6:1 and slighly zoomed in to avoid borders then both are indeed 16:10. It is the dvd AR that is then fooling you.

BTW What film is this and then we can check the correct AR.

Now I did not check the edges of both initial frames to see if the is indeed info missing on the dvd*

As for the 1.888 not sure where this comes in to the calculations.One reserve I have is that the level of magnification is different.

*edit: Actually there is more information on the dvd. So the images are not identical in the first place

I'm not sure to have understand what you mean.
anyway the film is Buongiorno, Notte 2003.
The 1.88 is just the value I read in the image editing software, irfanview. Is the AR of the selection.
I think bigger is this selection area, more accurate is the result.
About they are not identical, I said it previously: "- Sources have different crop."
That's why I would like to use a couple of reference points as selection borders.
6. Well IMDB lists the AR at 1.85:1. So neither the DVD or your initial calculations are correct.

But 1.88:1 is close
7. Originally Posted by DB83
Well IMDB lists the AR at 1.85:1. So neither the DVD or your initial calculations are correct.

But 1.88:1 is close
uhm.. I'm going in confusion, that 1.88 AR I mentioned before is just the AR of the selection, it can vary..
anyway here's the final AR comparison:
https://www.dropbox.com/s/gl21qdoppc0ylss/ar%20comparison.png?dl=0

edit:..without considering black borders crops.
8. Like I said, and now more than once, the confusion is arising due to re-framing.

Assuming that IMDB is correct then you have to end with an active AR of 1.85:1. How you achieve that is up to you. (rather late in the day now for me to consider the maths nut others might chip in)

Now you have both images in view you should clearly see that information is not consistant. The end result does not look good since it is 16:10 (1.6:1) not 18.5:10. Heads are now much taller.
9. uhm... I'm quite lost now..
I thought it was enough to match just a part of the image, like in example below...
I've changed selection, and again the AR is similar.
well, I did it with not great accuracy but this is just a test.

https://www.dropbox.com/s/1du31dbwgnkkoiy/ar%20selection%20comparison.png?dl=0
10. I need to sleep on this. To try to estalish how these images should have looked. Yet without the familiarity of the actors who am I to judge (the last film I watched from this director featured an actress who wore considerably less )

Of course you are assuming that you have the EXACT same frame in both versions.

I am not Italian. Yet some years ago I did watch a lot of Italian tv so I would have thought they were quite competant in their presentations. Did you ever consider that the WEBDL presentation was accurate (not AR wise but image-wise)
11. Universal formula for AR:

DAR = PAR * Horiz / Vert

And if comparing 2 images, DAR1 = PAR1 * Horiz1 / Vert1, and DAR2 = PAR2 * Horiz2 / Vert2.
Solve by equating one of the elements on the 1 side with its correspondent on the 2 side.

Note that this includes the full image, which is active picture plus any padding (letterboxing, pillarboxing), so it may be more complicated with real world examples, especially if your intention is to just retain the active picture.

Plus, as was already mentioned, it only works if one assumes they have the same framing.

Scott
12. Originally Posted by maudit
this can be considered a good and reliable method, or am I forgetting something?
Since the framing is different, then measuring distances between objects and calculating ratios is the only way I can think of.
Originally Posted by maudit
I suppose that higher is the distance between the two reference points, much more is the accuracy, right?
Yep.
Originally Posted by maudit
Can you suggest another, more precise, reliable or simpler method in order to achieve the objective?
No.

First, despite that there are black borders on the sides the DVD seems to have PAR 64∶45 as you said because there are black bars on top and bottom. So, the whole DVD screenshot seems to be true 16:9 to me. Therefore, I don't care about DVD PAR and can use square pixels from the screenshot.

A rectangle between man's and woman's eyes:

DVD: 430x168
WEBDL: 850x298

Calculate required horizontal distance between the eyes for WEBDL:
430*298/168 = 763

Calculate reduction factor:
763/850 = 0.8976

Calculate new width:
1920*0.8976 = 1723.5

My calculations gave me slightly narrower frame than yours. But visually comparing the two, it seems narrower than it should be, so I'd say that yours is about right.

But why do you want to bother with the WEBDL? To me, it has inferior quality compared to DVD, and it has a stupid channel bug. What is the name of the film? I think I know this actress.
13. Originally Posted by ConsumerDV
Originally Posted by maudit
this can be considered a good and reliable method, or am I forgetting something?
Since the framing is different, then measuring distances between objects and calculating ratios is the only way I can think of.
Originally Posted by maudit
I suppose that higher is the distance between the two reference points, much more is the accuracy, right?
Yep.
Originally Posted by maudit
Can you suggest another, more precise, reliable or simpler method in order to achieve the objective?
No.

First, despite that there are black borders on the sides the DVD seems to have PAR 64∶45 as you said because there are black bars on top and bottom. So, the whole DVD screenshot seems to be true 16:9 to me. Therefore, I don't care about DVD PAR and can use square pixels from the screenshot.

A rectangle between man's and woman's eyes:

DVD: 430x168
WEBDL: 850x298

Calculate required horizontal distance between the eyes for WEBDL:
430*298/168 = 763

Calculate reduction factor:
763/850 = 0.8976

Calculate new width:
1920*0.8976 = 1723.5

My calculations gave me slightly narrower frame than yours. But visually comparing the two, it seems narrower than it should be, so I'd say that yours is about right.

But why do you want to bother with the WEBDL? To me, it has inferior quality compared to DVD, and it has a stupid channel bug. What is the name of the film? I think I know this actress.

Yes ConsumerDV, you're right, the WEBDL seems to have inferior quality, I think it depends on the fact it is just a bad upscale,
and it is too much blurred, and yes it has a logo, I'm not 100% sure I'll release it, anyway it is the unique 1080p available.
This thing can happens again, so I have to be prepared for further cases, let's say is a kind of exercise.

The title is written above in a previous post, it's "Buongiorno, notte (Bellocchio, 2003)", the actress name's Maya Sansa.
14. As DB83 mentioned, your DVD version may not have a correct AR too, The only way of getting the right aspect ratio for the film is to look for a photograph of the actor and work the head proportions, it is sort of a forensic work but it is the only way to get a solid reference.
15. Like you originally stated, you have the dvd so you know the AR. It is not 1.85 as stated by IMDB. It is not 1.6:1 per your calculations. You are, again if the dvd cover is accurate, quite close since that is reported to be 1.66:1

So I also took the 1080p version in to a photo editor and resized it to 1792*1080 (rounded to the nearest mod4). The logic in doing this was to allow for the extra detail top and bottom in this. I then reframed the resize back to 1920*1080. Now you see the resultant border most of which vanishes when the dvd frame is zoomed in removing that additional detail on the web version.

So does this look more representative to you now we are using the actual AR.
16. I can't get the dropbox links to work, but if you're an Avisynth user, CropResize might make it easier.

You'd open the DVD version like this (the four zeros are placeholders for any cropping that might be required later on):

CropResize(1024,576, 0,0,-0,-0, InDAR=16.0/9.0) # 1024x576 or any 16:9 dimensions
or
CropResize(1024,576, 0,0,-0,-0, InSAR=64.0/45.0) # 1024x576 or any 16:9 dimensions

For the web-dl, I'm simply using the dimensions DB83 settled on as the input display aspect ratio and telling the script to add borders as required for a 16:9 frame (so with the borders it'll have the same dimensions/aspect ratio as the DVD).

CropResize(1920,1080, 0,0,-0,-0, InDAR=1792.0/1080.0, Borders=true)

The way I'd try to work out the correct InDAR is to create a script for each video, open both scripts with MPC-HC and run them maximised while switching between the two instances of MPC-HC to check if objects have the same shape in both videos. If they don't, you can fiddle with the InDAR for the web-dl and reload the script. InDAR=1795.5/1080.0 or whatever is required.

CropResize bases it's calculations on the specified InDAR so the output dimensions don't really matter as such, although it'd be easier to stick to 16:9 dimensions for both versions until you're happy they're correct.

If the two versions were cropped/framed differently they won't match perfectly in respect to the size of objects, but you can always apply some cropping with CropResize to zoom, or you can use the 1 and 9 keys on the numeric keypad to zoom one of the videos in or out a little in MPC-HC. The 5 key resets the video zoom.

As CropResize bases it's calculations on the InDAR or InSAR you tell it to use and won't distort the picture (assuming it's the correct InDAR/InSAR), all you need to adjust is the InDAR to correct the display aspect ratio. Once you're happy with the aspect ratio you can use CropResize to crop any existing black and/or add borders if need be. The shape of objects in the picture won't change... even if you change the output dimensions. CropResize will crop extra or add borders to compensate if it has to, but the shape of objects in the picture won't change as long as the same InDAR or InSAR is used.
17. Originally Posted by dellsam34
The only way of getting the right aspect ratio for the film is to look for a photograph of the actor and work the head proportions
Wow. Specifically this is the only way?
18. No, you just need some sort of known reference. For example a straight-on shot of an analog clock, which SHOULD be a true circle. Or some other known item, that is not skewed in the shot's perspective, but has a clearly defined, universally known, set AR (famous architecture is good).

This all assumes the the directors didn't INTEND to shoot with a changed AR.

You are at the mercy of what is made available in the clip(s), though.

Scott
19. Originally Posted by [ss]vegeta
Originally Posted by dellsam34
The only way of getting the right aspect ratio for the film is to look for a photograph of the actor and work the head proportions
Wow. Specifically this is the only way?
Knowing a perfectly circular or square object is not present in the movie as mentioned previously pretty much the only way, Should have said "another way" though.
20. It would be nice if the OP came back to this topic and commented on posts after his 'fix'

Of course I am biased in my own solution. But 60-70 pixels difference is a lot.
21. Originally Posted by DB83
It would be nice if the OP came back to this topic and commented on posts after his 'fix'

Of course I am biased in my own solution. But 60-70 pixels difference is a lot.

Hi, sorry for the delay, all is becoming quite hard for my poor skills.
I've tried on some different frames and I have calculated two possible choices:

(1728,1080) or (1732,1080)

probably 1732 is better but I'm not sure.

No way to find any spheres, here's some test frames:
https://www.dropbox.com/sh/nt2qa3ndkz3knj8/AAB1c0ud3jz-SPkhtwtaK3mUa?dl=0
22. What it comes down to is if you think the faces of the two actors look as natural in the resized webdl as they did in the original dvd.

You will appreciate that film AR is rarely 1.78:1, the displayed AR on a 16:9 dvd. So some re-framing is done.

But the problem persists that you are not comparing identically framed images so judgement in skewed.

In the new series of images you do have another point of reference - the 'O' in the title screen. What you might like to try is to take that image from the dvd, resize it to 1920*1080 and then draw a frame around it like you originally did. Note the AR so defined and compare that with the same excercise on the webdl 'O'. If you still end with an effective 1732*1080 then there is little more I can add.

But I would still ask you to visually compare the initial image from the dvd,resized to 1920*1080, both with your 1732*1080 version and the 1792*1080 version.
23. If that thing is supposed to be a circle the frame needs to be resized to approximatily 976x576:

[Attachment 64912 - Click to enlarge]
24. ^^Hmmn.

Never considered that one. But...

That AR is 1.69:1 which is closer to the suggested AR of the film (1.66:1) than the 1.60:1 achieved by the OP
25. If you are trying to choose between 1728 vs 1732, that is only a 0.2% difference. Not really enough to worry about (most people would never be able to tell the difference).

Scott
26. Originally Posted by DB83
^^Hmmn.

Never considered that one. But...

That AR is 1.69:1 which is closer to the suggested AR of the film (1.66:1) than the 1.60:1 achieved by the OP
I don't think there's any reason to believe the DVD has the correct aspect ratio. There are small black bars at the left and right of the frame and the video isn't very sharp -- indicating it was digitized from video tape, not directly from the film. As we all know, most often a 720x576 ITU cap is written directly to DVD without adjusting for the difference in aspect ratio between DVD MPEG 2 (64:45 SAR) and the ITU spec (~16:11 SAR). But the black bars aren't big enough for a standard ITU cap. So it has likely been manipulated a bit.

The circle (if indeed it is supposed to be a circle) in that particular frame isn't very big so my resizing could be off by a pixel or two.
27. ^^ But this is the issue with these two versions. One has more active information top and bottom whereas the other has more active information left snd right.

My only source of reference was the dvd cover. Could that be wrong ? sure. Could the IMDB info be wrong ? sure.

So even with the best intentions the only 'accurate' reference is one's own eyes. Yet many do not have perfect "20/20" vision. Some time back there was a topic that illustrated that 'perfect' circle with another. And several correspondents (myself included - but I could have been a minority of one) chose the wrong sample of 'perfection'.

The OP must decide (even if we disagree) with what looks right - and be prepared for the vitrol should he share his vision with others who hold a different opinion.
28. I didn't rely on my eyes. I opened the frame in an image editor and use the tools to count the pixels across the width and height of the circle, resized the full frame accordingly, then checked the circle in the resized image. A 270x270 crop of the downscaled frame:

[Attachment 64914 - Click to enlarge]

The OP did a good job of adjusting the webdl to the dvd. But his assumption that the DVD has the exact right aspect ratio is almost certainly wrong. Of course, he's free to do whatever he likes.

Statistics