Star Wars GOUT in HD using super resolution algorithm (* unfinished project *) — Page 19

@ Laserdisc Master

I certainly agree combining different methods has the best chance to achieve superior results. In fact my current version of "super resolution" now uses multiple other filters to enhance the results further. One of the great things about your algorithm is that it is able to greatly enhance SD material by using custom build methodologies. I think this is very important, because I believe there's a limit to what sharpening and denoising alone can achieve. Your upscales of the LD are a good example. The enhancement is astonishing, and I'm very sure you could not have gotten that far with sharpening and denoising alone. 

@ zee944

It's a bit ironic that it was Team Blu member You_Too who first suggested using the Spline64Resize as a benchmark for the super resolution upscale in this thread. Not that it is anywhere near the quality of their upscale, but apparently it's a reasonable enough choice for a comparison.  

@ DrDre,

The problem is that for this methodology no sharpening has been applied.

Maybe it wasn't called "sharpening", but it has been sharpened. If it's not sharpened, sharpen it. It will look oversharpened. Why? Beause it's already sharpened.

The fact that another method may be able to also retrieve these details doesn't take away from this fact.

Does that mean you acknowledge that a traditional denoising and sharpening can achieve (virtually) the same improvement? If it does, there's no point arguing any more, since that's what I was saying from the very beginning.

My problem was that you're happy to compare the results to a simple upscale, but struggle to compare it to more refined methods. If you're so sure of the superiority of superresolution, why's that? But if you're not, it's fine.

Now if you want to argue that super resolution has very little to offer, and that it is mostly due to sharpening and denoising

That was practically my first sentence in this thread...

It's been scientifically proven time and time again that super resolution can retrieve details better than most other methods.

Yeah, on very specific materials. Or maybe NASA does have something that works fine. For us, average humans, it's proven when you can get the thing yourself (the device or the software), take a perfectly average material, run it through, and end up with a result that is impossible to achieve with traditional methods. I've yet to see that happen. At least once... all I and everyone on the video processing forums have seen so far is just papers, theories, and magical results with specific materials noone could duplicate on other sources.

With that being said, there can be gains from superresolution here and there, but very little overall. I really wish it would improve videos significantly, I really wish... I have plenty of materials I could use it on. If only I would be proven wrong!...

You argue my comparison in my last post to the Avisynth Spline64Resize is unfair, but you forgot to mention that I also compared it to the true HD frame. This to me seems like the fairest comparison of all.

I didn't forgot, I was focusing on the unfair comparsion only. The other is fair in its method, but doesn't prove anything. The SW Blu-rays aren't very detailed unfortunately, and your processed frame is oversharpened, so you can't really compare them.

@ _,,,^..^,,,_,

Can you tell me what should I see? On the forum link, I can't find description about your method, nor the screencaps of The Abyss, The Thing etc. the posters talk about. Is it about image registration and merging?

I've taken a look at the SW comparsion. Do you say it's as big improvement as I achieved on the Twin Dragons caps, or better/worse?

I think the original idea was to merely experiment with Super Resolution and see the results. At no point was the intention of putting down Team Blu's work. I for one, would love to see some comparison screenshots of SR and Team Blu's upscale.

It should not be a competition, but a collaborative effort to fine tune, and perfect the enhancement process! :) 

@ zee944

No it does not mean I say sharpening and denoising can achieve the same results. For your information, the super resolution algorithm uses deblurring, not sharpening. There is a difference. You could have known this if you would have taken the time to read some of the scientific literature you dismiss. Again you offer nothing more than unsubstantiated opinions, and misrepresent what I have stated. If you would have taken the time to follow the link Laserdisc Master posted, you would have seen that there is nothing magical about super resolution, and that it's performance has been validated on a large number of real video samples. However, you seem to be interested only in throwing around baseless accusations. As I said put your money where your mouth is, and offer some real evidence, rather than hearsay. 

Edit:

As far as your Twin Dragons example is concerned, some details are enhanced, others are removed. Is it improved? Definitely. How does it compare to super resolution? No idea, since you're comparing apples and oranges. You can't possibly expect anyone to form an opinion based on two samples from a totally different source. 

@ Zyrother

Thank you! I could not agree more. 

Here's an interesting link from last year:

http://www.imaging-resource.com/news/2014/05/09/apple-awarded-iphone-patent-on-super-resolution-camera-system

If only they would have known about sharpening and denoising. That would have saved them millions of dollars developing some inferior super resolution camera system. 

O, I'm being sarcastic... :-P

@ DrDre,

I fearfully ask, doesn't deblurring make the image sharper?

I never intended to mispresent anything you've stated. But you do it all the time, and talk about things that are either evident or irrelevant.

I didn't dismiss the scientific literature. I even said SR works to a (small) extent. What you fail to understand that it doesn't matter what is possible theoretically or with inaccessible devices. You can only work with what is available, and you can't gain much with that.

Another thing you don't want to understand, that, say, 80% of your gains can be duplicated with denoising and sharpening. You don't want to hear about that. You act like every "new" detail is because of the substantial part of SuperResolution. That's wrong and harmful too.

And the third thing you don't understand that it is not personal and I have no conflicting interest with SuperResolution. I have plenty of footage I would love to improve with a well working method. But crediting SuperResolution for getting details you can also get with denoising and sharpening... that's not about finding out what your SuperResolution technique is capable of. That's something else.

I've seen the site Laserdisc Master posted. It's old stuff.

Comparing apples and oranges? Remember your words when you're comparing a simple resize to SuperResolution...

@ zee944

It's becoming tedious, but again only opinion, and nothing to back it up. You're 80% claim would mean something if you could provide some evidence. There is plenty of evidence to the contrary, and it's neither theoretical or inaccessible. I've got more faith in years of scientific research than your baseless claims. But like Laserdisc Master suggested, if you think you can do better, start a thread and prove your point, but until that time: "bachelor's wifes and maiden's children are well trained".

In any case, you're happy with your method, I'm happy with mine. Let's leave it at that. This discussion is pretty pointless anyway, since we're obviously never going to agree.

Edit: 

To answer your question: sharpening refers to the process of increasing contrast in edges of objects; deblurring refers to the process of deconvolution, which corrects a defect. To say it bluntly, sharpening treats the symptoms, deblurring treats the cause. It comes as no surprise the latter is generally preferred. 

@ Laserdisc Master

Wise words my friend...

Each upscale method is as unique as the source material it is trying to enhance. Some methods will work better for some material, others will not. I doubt there will ever be a universal upscale method that brings peace to the world.

The challenge is to determine what method works best with the source material. SR has so far, flexed its muscles, and proven it is very much more than capable of producing fantastic results. But, it can, and will always be improved through collaborative efforts, and constructive criticism.

@ Zyrother

And it has been and will be thanks to constructive critisism from forum members like yourself. 

Despite the recent attempted aggressive turn in this thread, I am impressed with everyone else maintaining their previous good natures.

Here's a comparison with a frame from Team Blu's upscale found on this forum.

Avisynth Spline64Resize vs Team Blu

http://screenshotcomparison.com/comparison/125656

Team Blu vs super resolution v8

http://screenshotcomparison.com/comparison/125657

Avisynth Spline64Resize vs super resolution v8

http://screenshotcomparison.com/comparison/125658

Ignoring the color differences of course. That's a very close comparison. First impressions is that SR V8 has more of the natural grain field left intact, which is a plus. But has slightly worse aliasing than Team Blu.

A couple more direct comparison screenshots between Team Blu and SR V8 would allow for a better evaluation!

My two cents:

Not surprisingly super resolution has more detail (openings in the stormtroopers helmets), but also more aliasing. There's also an upscaling artifact on one of the stormtroopers chests (difficult to spot). 

@ DrDre

There's no evidence without fair comparsions with some refined, but traditional methods. That's what I suggested in my first post: to compare your work to appropriate projects, and not only to a simple resize. I'm not the only who requested it. I, personally, would be curious to see what extra details you achieved via SR! If it's a lot, I will be excited. If it's just a little bit, well... that's what I expect. And I never said it would be your fault. It would be the fault of the accessible SR methods.

I was never talking about myself or my "method". I've only posted a few caps to show that the [illusion of] resolution improvement can be easily there without SR. Once it is accepted, how can you tell where the sharpening/denoising ends, and where the superresolution begins? You're experimenting SR because you think it can churn out more from the GOUT than the usual methods, right? Then wouldn't it be important for you to find out how much the "more" is?

I've tried the summarize what I requested from the very beginning as politely as I can. I hope it's not misunderstood in any way.

PS: I wasn't asking the difference there... I was trying to point out (sarcasictally) that deblurring makes the image sharper.

Luckily, the artifacts can be relatively easily avoided:

Team Blu vs super resolution v8

http://screenshotcomparison.com/comparison/125665

Avisynth Spline64Resize vs super resolution v8

http://screenshotcomparison.com/comparison/125667