Is there an advantage to HD displays for SD DVDs? (1 Viewer)

[ChrisWiggles]

Okay. I'll be very careful....?

No, I do not want to do the math, it would be a waste of my time. The math has already been done on several images that I have already posted on this thread. That's far more convincing, in my opinion, despite another user who doesn't seem to believe that Poynton knows what he's talking about.

All I see is that you're looking for a "gotcha" moment so you can continue pretending that basic sampling theory is wrong. Good luck with that.

 

[Jeff Gatie]

I never said the theory was wrong. I said that in real world conditions the theory can exacerbate some noise rather than decrease it. The Gaussian blur uses a bell curve analysis to assign intermediate values to the points surrounding the point being analyzed, thus "blurring" sharp edges and causing a more gradual transition from color to color, light to dark, etc. This is great with clean, sharp images and gives a significant benefit when used to scale an image like the one you linked, especially when displayed as the "before" image is - close enough to fully delineate the block structure. However, Gaussian blurs and other similar techniques tend to spread out the points that are anomalies in the image (such as my "speck" of white in an all black square) because the bell curve is now asked to cross a larger range of values. Depending on the size of the analysis matrix, this "spreading out" can extend to 1, 2, 4 or more points from the center before transitioning back to the original shade/color/brightness.

So, as I said, upscaling is a compromise between a better picture structure that can be viewed at closer distances vs. possible exacerbation of noise and a softer picture.

I'll have more as soon as I find an algorithm that translates easily without too much high level math (I'm a little rusty . . .

).

 

[ChrisWiggles]

Well, as I said a long time ago in this thread, you are better resolving an image. This includes resolving fine detail, noise, compression artifacts, grain etc. Both good and bad things. As I said before, I am interested in an accurate image. There are many ways we can obscure detail and noise, but that doesn't lead to an accurate image. If the image is noisy or a poor source, that's a source issue, and fully resolving the image is just presenting what is there both good and bad. With good content like a clean DVD there's a lot more to appreciate as things are better resolved.

 

[Jeff Gatie]

Ahhh, so we agree that upscaling can exacerbate the analog noise present in broadcast SD content? What took you so long? Seriously, that's all I was trying to say Chris. If you read back without the mindset that I couldn't comprehend stuff that was elementary to me 14 years ago, I think you will see that we agreed all along. I just don't get into the high falutin' theoretical classroom BS anymore. As I've said before - I'm an engineer. I build shit for a living.

 

[Fredrik L]

 

[Jeff Gatie]

Then why did he keep arguing?

 

[ChrisWiggles]

Because you made numerous statements that asserted that upscaling degrades the image. Well, that's simply not true. It's not I who has been argumentative or changing my positions in this thread. I've been trying to explain the same fundamentals in a way that you will accept or understand. You hold some misconceptions about video and what samples represent and how they should be reconstructed. There are better and worse ways to reconstruct an image. I'm explaining to you why one way is generally superior to another, but you don't seem to believe me. Now you're moving back into a position about "noise" in the content which is a distinct question that really is a side issue and something that as Frederik rightly points out I mentioned long long ago. Obviously, if there is noise in the content, and you are better resolving the content, you will also be better-resolving that noise for better and for worse. If you have crappy content, it's going to look crappy no matter what you try to do with it. This isn't news to anyone, and I've said as much several times in this thread. But again, to make crystal clear, that's not what interests me. What interests me is getting the most accurate image from any content, good or bad.

And I'm glad you're an engineer, but clearly you should do a little bit of reading or at least have some open-ness when dealing with a different field of engineering that you may not be as familiar with. I've suggested some texts you can look at, and I've gone into WAY more repetitive detail than I think is really necessary due to your stubbornness. I don't think it's fair to continue berating me with what you personally feel to be true. I've provided some images which I feel are pretty convincing as well to illustrate a little bit further by way of example what may be a little bit theoretically abstract.

Of course, at this point now I expect one of you to return to some other ridiculous point about how far away we are thus this doesn't make a difference. Well it's true, if you look at the images I posted and then move across the room from your monitor or display, eventually they will be indistinguishable. But I've covered that before as well. I'm not really sure what you expect me to say beyond what I've said before. If you expect me to cave in and admit that 2+2=5, well I'm sorry it's not going to happen.

 

[David_p_S]

The images that you have posted, Chris, do not prove your point. Each of the images are the exact same fixed pixel display size, each with the same pixel dimensions (that of my screen), and each with the same number of display pixels.

The difference in image quality is not because of different scaling, they were all scaled to the same display size!! The difference in image quality is due to better algorithms!

So what you're saying is, with a fixed display pixel size and count (like a tv image), an upscaled continuous tone image will look better after the upscale process if a better algorithm is used.

 

[Jeff Gatie]

And if that algorithm is optimized for pristine content such as reference quality DVD's, the performance of the upscaling on content that has inherent analog noise may be to severely degrade, not improve, the picture. Chris's only reply to this is he keeps telling me that I'm ignorant of the processes and/or the theory.

Chris, the very nature of the upscaling algorithm you gave an example of does not just make a crappy picture equally crappy, it makes it crappier. This is the notion that you will not concede. I have offered to prove it mathematically and you refused to look at it; and yet you persist in denigrating me with your thinly veiled suggestions of my ignorance - and I don't appreciate it. Upscaling algorithms assume a pristine source. They rely on smooth transitions. They do not do well with anomalies or things like perfectly vertical or horizontal lines. With anomalies, they tend to spread them out, making them blotchy and far more prevalent than they were in the original source. With sharply defined vertical or horizontal lines, they tend to blur the edges and leave them less defined. This is one of the weaknesses of upscaling (call it one of the compromises). My only point is that upscaling is great for what it does, but it does not always improve a picture - it is a compromise, certainly the best we have. However, I'd never tell a person that watches 90% broadcast SD who gets a fixed pixel HD display and watches it at SD type distances to expect a "far superior" picture to an ED set. He'll get an inferior picture. But that's OK, Chris. Really it is. You can reply with another statement of my ignorance, I will give another offer to prove it mathematically, you can refuse to accept it (oh the irony ) and we can go round and round. Frankly, I'm a little tired of it. I'll check out the Poynton book. It's just too bad you are not as open to a little math/computer science education yourself.

 

[ChrisWiggles]

I'm waiting for you to present any information at all. I've not seen any arguments or facts that bolster your position in any way. If my thinkly veiled suggestions that you don't know what you're talking about bother you, well, maybe my suggestions are not without merit. There are arguments that you could be making which are significant and relevant, but I've not seen you make any of them. For instance, there is the aliasing that is introduced when you're scaling to a resolution that is not many times greater than the original resolution and is not an even multiple. I've not seen you mention that as a possible drawback. You could have, but you didn't.

Go ahead and present an argument if you wish, or use some illustrations from a source that disagrees with the sources I've cited here. But don't pretend that I'm ignoring or dismissing preemptively what you might post, because that's not true and you know it. All I said is that I'm not going to waste my time doing a bunch of math here for you. You're free to do whatever you want. Post some images, post your interpretation of theory, etc. I've not seen any professionals highly regarded or otherwise cited on your behalf by you or anyone else. I suggest maybe you might want to start there.

 

[Jeff Gatie]

Maybe you should start by rereading the post where you think I suggested you do the math:


I asked if you wanted the math. I didn't ask if you wanted to do it. Contrary to your condescending attitude (and this is just one example of how you misinterpret my posts), I am quite capable of "doing the math."

More to come later.

By the way, any algorithms that you would prefer I illustrate? Also, I'll not be posting links to photos. How exactly can we see the effects of upscaling when the photos we see are rescaled to yours, mine and other's personal display settings? Sure I could take the route of your example and post it at a resolution that makes the block pattern 10 times larger than needed, but that is not anything but a finely tailored example to demonstrate the theory (a theory I agree with). Instead, I'll be dealing strictly with matrices and point values (hope that is not too "theoretical").

 

[David_p_S]

Any comment on this Chris? Do you agree that the pictures you posted only differ in upscale algorithm? (not sample count or pixel count or pixel size)

 

[ChrisWiggles]

I didn't fully understand your original comment. Can you clarify what you mean?

I understand that there are additional limitations imposed by the fact that these images have to be displayed and that's not constant. That's why they are "zoomed in" so as to be more exaggerated, they'll look basically the same on any reasonably average display system or computer monitor.

 

[ChrisWiggles]

They are grossly enlarged, such as presented by the photos I posted. This reduces or practically eliminates display concerns among individual systems. Come on, stop jerking people around. You really think your display is a serious limitations with the images I posted?

 

[Jeff Gatie]

No, I think the gross enlargement exaggerates the difference you would see at normal viewing distances. But then again, you said that doesn't count because you personally don't view at those distances, you view at a distance which allows you to see the picture structure (why, I don't know). I personally don't like to see the individual elements when I view my screen, but hey to each his own. I guess that is par for the course when hundreds of posts by people who don't like the upscaling of analog SD on their new HD displays is trumped by the anecdotal evidence of one person. But, I guess you make the rules here - I'll go play in the math pool, thanks.

Just give me a little time, I've got a major project I'm trying to drive to completion and as I said, my linear algebra is a little rusty. However, if you'd like to argue random number certification algorithms, I've got those right in the front of my mind.

 

[ChrisWiggles]

I sit at a very reasonable viewing distance usually 1.5-2.0x Screen widths. Most people in an HT environment sit at these kinds of distances. How far away do you sit? A half mile away from a 27inch direct view?

 

[Jeff Gatie]

I certainly don't sit at a distance that allows me to discern the individual picture elements. I like a filmlike experience. If I wanted to see a bunch of pixels, I'd look at examples from my old textbooks. Tell me, when you go to the theater, do you sit close enough to distinguish the individual film grains?
And again, I never said there was no purpose or benefits to upscaling. I said there are compromises. Please stop putting words in my mouth. Upscaling is great at what it does. There are limitations and compromises to upscaling, though. If you knew enough about the science behind the theories you spout, you'd see that.

 

[ChrisWiggles]

Go read the book I suggested, then come back and talk to me about my ignorance.

 

[Jeff Gatie]

Chris, show me the post where I said upscaling has no purpose and/or no benefit. Include the whole post if you quote it. Make sure you include the caveats in each statement I make. You sure have a habit of taking my statements that have detailed, specified conditions and turning them around to be general statements that only a moron would make.

Edit - By the way:


Do you realize the irony of this statement????

 

[ChrisWiggles]

The statement is not ironic at all. SD is not HD, let alone a good film print, and as a consequence it lacks the detail to create a fully engaging experience at very close angles. This should not surprise anyone.

I already addressed your apparent dismissal of upscaling in post #47. Early in the thread you made some contradictory statements, first you seemed to reject the idea that upscaling was beneficial, then made statements that "Not all scaling is alike and not all is beneficial." That's reasonable, but they don't mesh with your ideas about scaling elsewhere.

If you want to boil down my basic point, it's that a 1:1 sample to rectangular sample mapping of any content is among the worst ways to render that image. Conversely, the best way to render any image is to display that image at a theoretically infinite (realistically a "very high") resolution via high quality scaling, ideally sinc-type scaling (this is theoretically ideal). Neither of these two points are controversial when it comes to continuous tone images which is what we're dealing with here.

My only purpose in going to such great lengths in these explanations was to illustrate in a theoretical way *why* upscaling is beneficial. Many people have experience with the benefits that upscaling can provide, but they are stuck on the "how can it look better since we're limited by the source resolution?" idea, and this is linked to how people improperly understand pixels or samples. All I've tried to do was shed some light on this topic so that people can understand properly just why upscaling is capable of providing these benefits despite the impossibility of ever increasing the resolution beyond what we are given to work with.

Unfortunately my efforts have been received poorly and with some disdain and anger. I guess I shouldn't feel slighted or insulted by this, yet somehow I do. I suppose in the future I should just refrain from trying to add the understanding of the forum? People can keep thinking with squares and end up with poorer quality images than they could have had. I suppose my generosity is unwanted and I should just not help anyone ever. We should never share our knowledge or help others. I should just get the best from my system and not help anyone else with theirs...?