What's new

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

Jeff Gatie

Senior HTF Member
Joined
Aug 19, 2002
Messages
6,531

Is where we part company. You cannot achieve the same results with a upscale to infinite resolution. You cannot believe that your can sit closer and closer to an SD image (zoom in), as long as the distance is one which does not fully resolve the picture elements; that as long as you upscale the picture it will always be superior. A good upscaling algorithm will be finely tailored to go from one particular resolution to another. It also will hopefully do it's job in one pass. To say this algorithm will scale to infinity and always yield a superior picture (given the caveats to what is "superior" - that is able to be view at a bigger angle of view, i.e. "zoom in" - that we now agree on) is also hooey. A Gaussian blur, used iteration after iteration, will soon reach a point of diminishing return and that point is well before infinity. Unless those CSI's really did take a low-res traffic camera photo, blow up a small reflection off an eyeglass lens and get a 5 megapixel mugshot.

Think about it, Chris. Really think about it. It really is the absurd CSI scenario above that your "infinite resolution" statement is alluding to.
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791

The same results as what? I am unclear what you're comparing to.

At this point I really mainly just see a reluctance on your part to accept what I've been saying in the manner I've characterized it. There is nothing controversial to what I'm saying, it really is just application of basic principles. And anyone with experience viewing DVD on high-resolution displays will quickly recognize that in this case the theory does very much translate into real and palpable benefits.
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791
I finally was able to find the thread I was searching for over in the AVS Archives. I wasn't able to find it previously because I was searching under the wrong thread-starter. Anyway, this has some more persuasive illustrations of exactly what I mean:

http://archive.avsforum.com/avs-vb/s...threadid=95576

"The best case would be to scale this with a sinc filter to an infinite resolution." Which is, in a nutshell, what I've been attempting to convince you of. Perhaps these images you will find more convincing, though they are no different really than the previous images from Poynton.
 

Jeff Gatie

Senior HTF Member
Joined
Aug 19, 2002
Messages
6,531

Chris, you have wavered in what you've said, therefore how can I know what I'm agreeing or disagreeing with? You started off by saying you can "recover" data from the original. We have since agreed this is not possible. You started off by saying the application of upscaling "always" reduces noise, but you did not reply to my real world offer to apply a Gaussian blur to a noisy (not blocky, noisy) real world photo to see if a Gaussian blur improves or exacerbates noise. You have backed off your original statment that the upscaled picture is always superior to agree with me that it is only "superior" if the goal is a larger angle of view. Now you keep saying that the upscale will yield the same "superior" results that we receive from the example above up to and including an "infinite resolution". Well, to me "superior" up to an infinite resolution means the effect that made the first upscale "superior" (i.e. we can veiw at a larger size (zoom) without the "noise" from blocking) will continue up to infinity. This is the same as saying we can zoom an image up to infinity and we will be able to recognize a superior image at each level of zoom, as long as we perform the Gaussian blur. This is decidedly not true (you yourself have said there is a diminishing return), so the "infinite resolution" statement is false.

Can we achieve a better picture from upscaling if we want to sit closer? Yes. Is it "better" than viewing the original at optimum distance (i.e. more detail, more sharpness, etc)? If the optimum distance is defined as close enough so as not to resolve individual pixels, then the answer is no, it is not better. That's all I've been trying to say and if you read what I wrote, Poynton agrees with me.

Edit: And after reading the thread you just linked, Bjoern also agrees with me. While the "infinite resolution" statement is a nice theoretical exercise, real world applications don't often adhere to the theoretical and even if they did, what good would it be to keep sitting closer and closer to an ever increasingly soft picture?;)
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791

It wouldn't do you any good. That's not the goal. You cannot increase the resolution in this manner, which apparently you do not grasp. You seem to think I am claiming that you can increase the resolution or "create" information beyond what we are provided by the content. I don't know how you can continue to think that this is what I am claiming despite my tireless and detailed explanations and explicit statements that this is not possible. My first detailed post in this thread stated unequivicolly that "You are not adding information...it is true that you cannot *add* information that is not present." And I have repeated that point several times, as does Bjoern. The goal is simply to reduce the introduction of noise at the viewing distances we are normally concerned with. To this end, upscaling to a higher resolution display provides significant, REAL benefits visible to anyone with average visual acuity. Using displays(and scaling to that display resolution) beyond the point that we cease to see display structure visibility provides no more benefits. But given that even 720p displays reveal structure within maybe 2.0 screen widths and the vast majority of HT viewers are well within 2.0 screen widths, the claim even that upscaling 720 provides no benefits is dubious, let alone the one that upscaling 480p content like DVD provides no benefits. You haven't been fully clear if that is, indeed, what you are asserting, but it seems possible that it is what you claim given some of your statements. Whether you are indeed claiming that or not, that claim is made by many people along the lines that "viewing at the source's native resolution is ideal" and that claim is clearly and unambiguously incorrect. That is really, fundamentally my point in this thread. Viewing at the source's native resolution is NOT ideal for the reasons I have attempted to explain here.

So to summarize in essence what Bjoern, Poynton, and myself are all saying, is that upscaling an image upon display provides benefits. Or in other words in response to the question:

"Is viewing a DVD at it's native resolution of 720x480 the best way to view a DVD?"

The answer is clearly NO.

Do you agree or disagree with this?
 

Jeff Gatie

Senior HTF Member
Joined
Aug 19, 2002
Messages
6,531
So now you are limiting your definition to just "aliasing noise". Ok, I agree with you. It will eliminate aliasing noise (duh).

Let's just agree to disagree, huh Chris? You can keep adding conditions to your initial statements and claim I misunderstood, if you like. I'm not going to argue with you. Let's me just leave you with this - An upscaled image of a noisy analog SD signal may eliminate "aliasing noise", but it will greatly exacerbate other forms of noise which are quite prevalent in the analog signal, far outweiging the benefits of upscaling. Therefore, in this particular case, watching SD on an HD display will be worse than watching it on an HD display. So the answer to your question about DVD is - I agree. Then again, the subject before was SD sources, not DVD. :frowning:

Edit - By the way, could you point out where I said upscaling provides no benefits?
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791

You stated, as I quoted in the last post that: "each interation of the scaling algorithm decreases detail and increases noise." I'm not sure how else one would interpret that kind of claim.

It does dissappoint me, however, that you refuse to engage the basic issues. In any case, there is little more I can say on this topic without treading on already well-covered terrain. I do strongly suggest you seek out Poynton's book if you're interested in exploring these issues and others in greater detail.
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791
Attached are three more images I dug up which are interesting and relevant.

One is unaltered
One has been blurred
One has been bicubic resampled


 

Kevin C Brown

Senior HTF Member
Joined
Aug 3, 2000
Messages
5,723
Chris- You mention not buying into Consumer's Reports display reviews. That's fine. But it's been reported here multiple times that they do have their sets ISF calibrated. Plus, their results usually do correlate with what the general impressions on AVS are in terms of display quality from different manufacturers and the technologies used. Plus other profressionally done reviews by HT Mag, WSR, CNet, S&V, etc.

Re: the earlier pictures you posted, "Each photo is identical, and is 16x20 pixels (in this case pixels is used synonomously with my use of "sample," so each image is the identical 16x20 samples reconstructed(displayed) with two different methods." I can see the pixels for each left hand picture. With a resolution of 16x20, I should also certainly be able to see the pixels for the righthand images and I can't. I even blew up the right hand pictures by 400%, and I *still* can't see the pixels! Pixels are pixels and upscaling doesn't change the fact that the only difference between resolutions is their size. The righthand resolutions are much higher. More than 4x even, which is approx 480p to 1080i/p.

I possible agree that theoretically, there are upsclaing algorythms that do a better job than simply turning 1 pixel into 4, or using simple interpolation between pixels in order to go from 1 to 4, but it's not even guarenteed which manufacturers use the better upscaling algorythms. And I keep going back to my own personal observations (in addition to others), that outside of a distance where the pixel structure itself is visible, I simply have not seen any benefit for SD on an ED display vs SD on an HD display, certainly not a "a very very significant benefit". The truth is in the viewing.
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791

I've said as much myself. By the same token higher-resolution images provide no benefit also. It is a logical fallacy to then claim that upscaling does not provide benefits. By the same token you'd have to say that higher resolution images provide no benefits. Well, under certain circumstances that is correct, but under others that's a ridiculous claim. Under normal HT viewing circumstance the claims that HD images provide no benefit over SD images is as equally ridiculous as the claims that upscaling provides no benefits.

And you're correct, the truth lies in the viewing. View the pictures. View upscaled content on a high-definition display. If you see no differences, I'm sorry but I really have to question whether you have any serious visual problems.
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791
You may also find these more in-depth discussions of why pixels do NOT have an area interesting:

"A Pixel Is Not A Little Square, A Pixel Is Not A Little Square, A Pixel Is Not A Little Square!"
ftp://ftp.alvyray.com/Acrobat/6_Pixel.pdf

and "A brief history of 'Pixel'"
http://www.foveon.com/files/ABriefHistoryofPixel2.pdf

Needless to say, a Pixel is NOT a little square... ;)

I know that a lot of people think of pixels as rectangular, but that is NOT true. Unfortunately this misconception persists. And it leads to discussions such as this one where people do not really understand what a sample is, and thus they do not understand what it means to "reconstruct" a waveform from samples (such as reconstructing an image).

BTW: wikipedia actually gets this right: http://en.wikipedia.org/wiki/Pixel
 

Kevin C Brown

Senior HTF Member
Joined
Aug 3, 2000
Messages
5,723
I am not going to argue with you anymore because it's becoming pointless.

The data is digital, which when it's stored, has no area. Agreed. But as soon as you display it, the pixels do have area, which has a direct influence on video quality. The size and number of pixels defines resolution.

I took the earlier pictures you posted. I cut and pasted one easily seen pixel from the left side to the right, and compared what I saw. Within the single pixel from the left side, I saw a singular uniform color (well, a shade of gray). When comparing that to the same size area on the picture on the right, know what I saw? A 10x10 matrix of different shades of gray. So the pictures are not correct. Or, they are not an indication of what happens when you go from 480p to 720p or 1080i/p. Which is really what this discussion is all about.


No, you keep quoting theoretical when the people who have actually done this kind of comparison all seem to disagree with you. SD on an ED display vs SD upscaled on an HD display. Outside of the distance where the pixel structure is visible. No HD content. It is not "a very very significant benefit" no matter how much verbosity you apply to your arguments.
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791

I've spent a very significant time looking at images upscaled on various kinds of displays in high-end environments, and it is beyond question that upscaling provides benefits which, however you want to define it semantically, are clearly "significant."

In response to this statement:
the resulting image would be EXACTLY the same when you use a native 720 x 480 panel or a 1920 x 1080 panel when the SOURCE only has the small resolution of 720 x 480.

Bjoern Roy puts it equally strongly:
No. Have you ever actually seen such things in real life? If you had, you certainly wouldn't claim this. Take an old lowend 7" CRT that can just resolve 480p with touching lines. Run an HTPC at 720x480 and watch the result. This represents a gausian blurred 720x480 case. Now take a perfectly focused 9" CRT and run it at 1920x1080 with a Radeon. The 9" CRT is going to look SO MUCH more 3D like and detailed, with the SAME 720x480 DVD source, its not funny.

Well I agree strongly with Bjoern. Anyone who has seen upscaled content in a quality system would have to be pretty blind not to see the differences. I've never had any video engineer tell me that they thought upscaling was undesireable or did not lead to significant improvements in the ability to fully reconstruct an image. As I've said before, it's a very simple 2-second comparison in person, but you're not in my theater so I can't illustrate in person. The best I can do is explain the theoretical reasons, and then give illustrations which are the best I can convey online. Of course we should all be skeptical just how much theory ends up correct in real practice, but in this case it's really quite clear cut. Anyone viewing, experienced or not, can see the differences.
 

Fredrik L

Agent
Joined
Aug 29, 2002
Messages
35
I find this discussion very interesting, and I appreciate all ChrisWiggles efforts to explain the details of samples and pixles. I'm not a technical person so the multiple explanations from different viewpoint is extremely helpful to get my head wrapped around this.

Anyway, I just have a simple question regarding the connection between upscaling and the theoretical proposition of its benefits:

Have I understood it right that the technique of upscaling a source to higher resolution always should be considered a gausian technique? In other words, is the sample distribution on an upscaled DVD-image (say 480p to 720p) always performed with the gausian effect? Is this an intended effect built into the upscalingchip or is it just a byproduct from the need to fit a smaller image to a larger canvas?
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791

It's not necessarily gaussian, actually it's usually not. There are many functions you can use, for instance in quality processing bicubic scaling functions are often used over bilinear scaling for instance, and there are others. Bicubic interpolation uses more samples than biliner, so it's more complex but better.
 

Kevin C Brown

Senior HTF Member
Joined
Aug 3, 2000
Messages
5,723
As soon as you put an image on a display, it *is* made up of little squares (or rectangles). Not samples.

Then somehow you posted the pictures incorrectly. Allow me to repeat: I take left and right images and blow them up 800%. I take the size of a *single* square pixel from the left hand picture, which does have a singular uniform shading of gray, and I overlay that same size pixel onto the right hand picture, I do *not* find a singluar shade of gray. It is a 10x10 square array occuping the same space. The resolutions of the lefthand and right hand pictures are not the same, no matter how times you insist that they are. If the resolutions were indeed the same, I shouldn't be able to see the 10x10 array of *different* shadings in the same space on the right hand picture as the left, where it is a singular, uniform shade.

And some examples, pg 3:

http://www.soundandvisionmag.com/hdt...-lcd-hdtv.html


"About average". Hmmm...

Go look up *any* review of an HD display where they critically look at SD quality. CNet, TPV, S&V, WSR, Consumers Reports, etc. Most of them will say the same thing. SD quality upscaled for an HD display is simply not that great. Most upscaling does nothing to improve on an SD source displayed on an ED display.
 

ChrisWiggles

Senior HTF Member
Joined
Aug 19, 2002
Messages
4,791

Well, you're simply wrong about that. If you can see an image, you are looking at a reconstructioni from samples. The two images presented are just two different examples of ways you can reconstruct the SAME 16x20 samples. Charles Poynton's text is a seminal text on digital video. If you think he doesn't know what he's talking about, you can feel free to email him and accuse him of being in error. I have merely presented, unedited, exactly what is presented in the book, and it looks exactly the same as it does on the printed page that I am looking at right now.

And I'm not particularly concerned what some magazine says about a random consumer display. Magazines get things wrong all the time, or they report on something that is observed and draw errant conclusions from that, or readers such as yourself reach your own errant conclusions.

The first quote has basically no relevance: "Also, while the set did pretty well upconverting clean 480i DVDs to fit its 1080p screen, its two noise-reduction modes weren't much help with poor-quality 480i cable broadcasts, which generally looked soft and noisy." This sounds to me like a source problem. SD cable broadcasts look soft and noisy in my environment as well. If you think that can be attributed to scaling operations, you are in error. If you think I would rather leave 480i as native 480i, you are also seriously mistaken, that would look very significantly worse, I'd be looking at hideous scanline and twitter and just horrible display structure on top of crappy content.

Second quote: "With standard-def sources delivered via the component-video input at 480i, the Sharp LC-46D62U delivered image quality that was about average. "

Likewise has almost no bearing on the discussion at all.

But as I said, gee, I'm biased to trust professionals who spend all their lives on this stuff rather than a random magazine reviewer I do not know. And anyway, the statements they are making are completely expected and fair. Crappy content isn't going to look good no matter what you do to it.

And as I suggested to Jeff above, go buy Poynton's text. The only way to understand what's going on is to, basically, understand what's going on in an in-depth manner. Arguing me irrationally in circles doesn't get you anywhere in that regard. Go read. Start with what I linked, the wikipedia entry on sampling, etc. Go buy Poynton's book. Then see if you can maintain the fantasy that samples are a bunch of rectangles. A point sample is a point by it's very definition. It really doesn't get more complicated than that. A point is not a square, or a triangle, or a circle, or aline. It's not a cube, or a parallelogram. It's not a blob of grape jelly. It's a point.
 

Jeff Gatie

Senior HTF Member
Joined
Aug 19, 2002
Messages
6,531
Chris,

Here is a picture with a grey scale of 10 (1 being black, 0 being white). The picture is of an all black square, denoted by a 10x5 matrix. Due to analog noise, it contains a speck - i.e. a single white sample in the picture:


1111111111
1111111111
1111111111
1111111011
1111111111


Now, let's scale it up (Note - I'm not going to do the math, just an example of what an algorithm might do. If you want the math, I'll get some real algorithms). In this case, we will represent the 10x5 as 20x10

11111111111111111111
11111111111111111111
11111111111111111111
11111111111111111111
11111111111111122111
11111111111112255211
11111111111335578752
11111111112255890985
11111111111225589552
11111111111133355331

What does the scaling algorithm do to the inherent noise in the original picture? Which picture is better when shown at the same size? Which best represents the original black square? What are the inherent strengths and weaknesses of this algorithm when dealing with analog noise?

Edit: Whoops, got my matrix sizes wrong.
 

Users who are viewing this thread

Forum statistics

Threads
356,814
Messages
5,123,666
Members
144,184
Latest member
H-508
Recent bookmarks
0
Top