Is an HD Super 8 transfer overkill?

[carllooper]

Because down scaling from HD usually produces better results than capturing in SD to begin with.

EXACTLY.

If the downscaling produced worse results it would be of no value in terms of what is being discussed, ie. the point at which "overkill" occurs.

To put it another way, if I resample at an even lower defintion - to better approximate a real world 1K transfer - then the difference from a 2K transfer would be even more pronounced.

Meaning that the "overkill" point must be set at an even higher resolution - ie. to account for real world deficiencys.

Carl

[Uppsala BildTeknik]

EXACTLY.

I really do not, at all, understand what you mean.

If you want to compare a SD transfer to a HD transfer, you need to use a SD transfer. Why is this so hard to understand?

If you play with upscaling and downscaling a image in whatever resolution, that is exactly what you do. You show what scaling can do to a specific image. But you are not, in fact, comparing a SD transfer to a HD transfer.

Simple as that.

[carllooper]

I really do not, at all, understand what you mean.

I can see that.

If you want to compare a SD transfer to a HD transfer, you need to use a SD transfer. Why is this so hard to understand?

You tell me.

If you play with upscaling and downscaling a image in whatever resolution, that is exactly what you do. You show what scaling can do to a specific image. But you are not, in fact, comparing a SD transfer to a HD transfer..

Yes, but I'm not doing anything much different from what you are doing. You are playing with upscaling and downscaling the original film. I'm playing with upscaling and downscaling a copy of the original film.

The purpose of both activitys is to find if the "overkill" point is within the limits of our scanning.

I haven't found it and neither have you.

Carl

[Uppsala BildTeknik]

Yes, but I'm not doing anything much different from what you are doing.

I´m comparing two different transfers to each other, trying to see which transfer will give better results. And I have come to the conclusion that a HD transfer will give you better results if compared to a SD transfer and viewed on a large HDTV.

You are looking at what scaling can do to a image. Nothing wrong with that. But it is not a comparison between a SD transfer and a HD transfer.

[MovieStuff]

The problem with this discussion is that the term "overkill" has not been defined.

Roger

[carllooper]

The problem with this discussion is that the term "overkill" has not been defined. Roger

Here is a defintion we can use.

Overkill occurs where the difference between two candadate scans, A and B, produces an image where the original signal is no longer discernible, ie. we see only noise.

For example, lets suppose scan A is 4K and scan B is 8K.

If we compare these two scans - which means computing the difference (as you might do using the Photoshop Layer Blend of the same name) - AND that all you can see is noise, then scan B is "overkill".

By how much is not known but what one will know is that it is somewhere below scan B.

The two scans I used were 2K and 1K.

And I was still getting a signal.

Here are the difference signals I get:

2K and 1K
http://members.iinet.net.au/~carllooper ... erence.jpg

2K and 0.5K
http://members.iinet.net.au/~carllooper ... rence2.jpg

Note that as you decrease the lower definition signal the difference signal gets stronger - further away from noise - so there is no point in using real world SD scans if the virtual SD scans are better. Real world SD scans just increase the difference signal - meaning you get a signal further away from where you hope to find the overkill threshold.

Carl

[Uppsala BildTeknik]

If we compare these two scans - which means computing the difference...

No it doesen´t mean computing, at least it doesen´t in my book. Here is a huge difference in how we compare things. I look at both transfers with my eyes, and then decide whatever I decide regarding the transfers.

If you cannot tell the difference with your eyes, if you are forced to do these kind of mathematical computing comparisons... Well I am curious about why?

Why does it matter, if you cannot see the difference with your eyes when you look at the end product? When you look at the moving images on a TV or a projection?

[carllooper]

No it doesen´t mean computing, at least it doesen´t in my book. Here is a huge difference in how we compare things. I look at both transfers with my eyes, and then decide whatever I decide regarding the transfers.

If you cannot tell the difference with your eyes, if you are forced to do these kind of mathematical computing comparisons... Well I am curious about why?

Well I'm not forced to do the computaions. The computaions are just another way to express what I'm otherwise computing with my eye/brain. As are these words.

Why does it matter, if you cannot see the difference with your eyes when you look at the end product? When you look at the moving images on a TV or a projection?

But I can see the difference with my eyes.

As can you.

Carl

[Uppsala BildTeknik]

As can you.

OK. Now I am curious to where I said I did. As far as I can recall, I said I can´t see any difference in image detail, other than the grain being softer on one of your images.

Where did I say I could see a difference in image details? Because it is image details that are interesting, correct? Or are you interested in getting as much and as sharp grain as possible from a transfer?

The way I understand it, from talking to people who have tested and compared 1080p transfers to 720p transfers, is that you are not getting more image details at 1080p. You are however getting more and sharper grain. This will, IMHO, just make it look worse (because more grain is not something I strive for). And this would most certainly be true for even higher resolutions than 1080p.

If there are no more details to gain from the person/landscape/thing that was in front of the camera lens at the time of film exposure, when scanning at a higher resolution, then there is no point in scanning at a higher resolution. A sharper and more pronounced grain structure is not more pleasing to my eyes, on the contrary.

And there is just no way you can convince me that a super8 camera lens from the -70s is capable of resolving a 3k image (or higher). That is just crazy talk and a pixelhunt for a higher pixelcount. For nothing.

[carllooper]

OK. Now I am curious to where I said I did. As far as I can recall, I said I can´t see any difference in image detail, other than the grain being softer on one of your images.

Where did I say I could see a difference in image details? Because it is image details that are interesting, correct? Or are you interested in getting as much and as sharp grain as possible from a transfer?

Ok. Fair enough. I can see that both the noise and the signal are softer. I thought that is what you meant as well.

The way I understand it, from talking to people who have tested and compared 1080p transfers to 720p transfers, is that you are not getting more image details at 1080p. You are however getting more and sharper grain.

Those people are wrong. The difference signal between 2K and 1K demonstrates that there is still a residual signal in the 2K sample. it hasn't completely reached the threshold where noise takes over.

This will, IMHO, just make it look worse (because more grain is not something I strive for). And this would most certainly be true for even higher resolutions than 1080p.

That's a fair enough comment. You can scan at a lower definition to correct for this particular discomfort. Bt there also other ways that do not involve lowering the spatial definition.

If there are no more details to gain from the person/landscape/thing that is in front of the camera lens, when scanning at a higher resolution, then there is no point in scanning at a higher resolution. A sharper and more pronounced grain structure is not more pleasing to my eyes, on the contrary.

But there are details. The difference signal is exactly that. An image of those details. That teapot (the thing in front of the camera) that you can see in the difference signal are the details. Here is an image of the details you would otherwise lose if scanning at the lower definition:

http://members.iinet.net.au/~carllooper ... erence.jpg

Where does this decernible teapot signal come from? From our imagination? No - it comes out of the 2K scan after the 1K scan has been subtracted. I didn't synthesise that signal. It's an emperical experimental eyeballable real world signal. Here it is again, in case you missed it the first time:

http://members.iinet.net.au/~carllooper ... erence.jpg

Indeed, the very fact that it is there in the data means that it can be used to computationally enhance the signal to noise ratio, ie. suppress the noise - which is what we do when watching a film with our eyes/brain anyway.

And there is just no way you can convince me that a super8 camera lens from the -70s is capable of resolving a 3k image (or higher). That is just crazy talk and a pixelhunt for a higher pixelcount. For nothing.

So call me crazy.

But look - if formal/mathematical reasoning is not enough - and emperical eyeball testing is not enough - and reasonable common sensible combinations of the two is not enough - then in what other philosophical framework do you suggest we continue this lively discussion?

Carl

[Uppsala BildTeknik]

I have no idea what you are talking about, again. If you can´t described the difference in image detail, as in tell me where you see details in the higher resolution that you cannot see in the lower resolution...

Then you can´t tell me the difference, can you?

Oh, you can show your nice little teapot how many times you want, it makes no difference. Unless you are going to look at those ugly images of a teapot all day long, what good are they for? They are of no use at all.

You can dream about super8 having 3k of resolution if that makes you feel better. There is no point in discussing with you when you don´t even understand why your comparisons are all wrong. They are not comparisons from a 1k transfer compared to a 2k transfer. Just playing with resizing a still image from A to B, and then back to A.

Go 3k, go! (or not)

[Uppsala BildTeknik]

I´m just curious, anyone else here think that 3k or 4k transfers from super8 is the way to go?

[carllooper]

I have no idea what you are talking about, again.

Which is why I continue to persist.

If you can´t described the difference in image detail, as in tell me where you see details in the higher resolution that you cannot see in the lower resolution... Then you can´t tell me the difference, can you?

I can. I've described the difference in a number of different ways: in words, mathematically, and with a precise image.

Oh, you can show your nice little teapot how many times you want, it makes no difference. Unless you are going to look at those ugly images of a teapot all day long, what good are they for? They are of no use at all.

I could say the same about your test chart as well but the issue is not about the artistic qualitys, or otherwise, of the test images. It is a technical discussion and any ugly image will do.

You can dream about super8 having 3k of resolution if that makes you feel better.

I'm not dreaming about 3K or anything else for that matter. You are dreaming if you can't see a real signal in the test I posted.

There is no point in discussing with you when you don´t even understand why your comparisons are all wrong.

Ok. So tell me. Tell me why my comparisons are all wrong. I'm open to argument. Really.

They are not comparisons from a 1k transfer compared to a 2k transfer. Just playing with resizing a still image from A to B, and then back to A.

So you're arguing that what you're doing is not resampling an image when you sample a film at different definitions.

But you are wrong. You are doing the same thing as me. The actual numbers, in terms of sampling size and various other minor factors, may vary a little - but in terms of what the debate is about - it's the same thing.

Good grief. I could just claim - like you - that I see no difference between the two images - even when the evidence is right there in front of me, albiet in an ugly teapot.

How do you expect me to agree with that?

I have this problem. I have this problem that when something is both emperically evident and logically, mathematically clear, that I can't help but feel it is closer to the truth than something that someone just syas is so.

Carl

[Uppsala BildTeknik]

Ok. So tell me. Tell me why my comparisons are all wrong.

It has been stated many times in this thread. Simply because a downscaled HD transfer will not look the same as a native SD transfer. If you use the HD transfer and downscale it to SD in order to compare, then the comparison will not be correct.

Because it will not show you what a SD transfer would look like. It will show you what a downscaled HD transfer would look like.

[carllooper]

Ok. So tell me. Tell me why my comparisons are all wrong.

It has been stated many times in this thread. Simply because a downscaled HD transfer will not look the same as a native SD transfer. If you use the HD transfer and downscale it to SD in order to compare, then the comparison will not be correct.

Because it will not show you what a SD transfer would look like. It will show you what a downscaled HD transfer would look like.

If I were to generate a range of downsampled images from my HD scan, from the HD scan sampling rate (3K) all the way down to a 1 pixel sample, then at least one of those downsamples must resemble an SD transfer in terms of the spatial resolution.

While it won't be exactly the same, the difference will be, I'd argue, far more difficult to see , with eyeballs, or computer, than the comparisons I've been posting.

You show me the proof that they would be different. You haven't done that. Can you post an SD frame and an HD frame of your test chart?

I'll resample your HD frame to resemble the SD frame and then you can tell me what the difference is.

Deal?

Or are you are scared I might be right.

Carl