more pixels are better!

Started Dec 14, 2008 | Discussions
John Sheehy Forum Pro • Posts: 26,227
Re: To be more specific...

DaSigmaGuy wrote:

Everyone knows that downsizing an image hides image noise so its
hardly supprising your technique of downsizing the larger 50D image
to the size of the 40D image makes it look less noisy!...Your
technique is therefore completely flawed and the results therefore
irrelevant.

Downsizing an image with proper resampling does not "hide" noise. Unless the noise was periodic, and the new pixel size an integer multiple of the period, the noise is still there. It exists at a smaller depth, in a bigger pixel, with the same effective noise at all common image frequencies with the original.

Take any noise from a camera with low MP, and upsize it to the number of pixels of current high-MP cameras, and the noise becomes more intense at the same resulting-pixel magnification. That's because the real issue in the visibility of noise is the displayed magnification of the image. You don't need to downsample to "hide" noise. You simply have to display the image smaller. If you have to "hide" noise from a 15MP to match an 8MP, then simply displaying them at the same size would also give equal visible noise at common image frequencies, without any loss of resolution for the 15MP. Imagine that your monitor was 100MP instead of 1 or 2 MP, and you had to upsample all images to view them full-screen, and that 100% pixel zoom viewing had no special interest, except to make higher-MP images larger than lower-MP images, but none filling the screen very much. Then, we could truly say that the lower-MP is HIDING its noise by being displayed at a smaller size.

Displaying images small hides noise. Downsampling does not hide noise; it only hides resolution.

-- hide signature --

John

ejmartin Veteran Member • Posts: 6,274
Re: To be more specific...

John Sheehy wrote:

Take any noise from a camera with low MP, and upsize it to the number
of pixels of current high-MP cameras, and the noise becomes more
intense at the same resulting-pixel magnification.

I'm not sure what you mean by this. Upsampling does not make the noise more intense at any fixed spatial frequency of the original image.

-- hide signature --
Iliah Borg Forum Pro • Posts: 29,167
Re: More pixels are not necessarily better...

The output image size is therefore determined purely by how many
pixels a sensor has

Totally wrong. The output image size is determined by the needs of the photographer/publisher and by the acceptable quality criteria. Also, it is determined by the time and money spent on preparing the image to print.

I have 3 cameras with the same pixel counts. I can't print to the same size from all of them, given the quality criteria and printing method are established the same for all three.

Worse, I have 2 cameras, one of which has lower pixel count but allows to print larger while I stay within certain underexposure (many call it ISO range).

Pixel noise is just the way to determine perceived image noise. Perceived image noise, like it is with film grain, depends on the print size. Abstracting from printing process (and/or display size) invalidates any noise discussion. This one is a good example.

-- hide signature --
John Sheehy Forum Pro • Posts: 26,227
Re: To be more specific...

ejmartin wrote:

John Sheehy wrote:

Take any noise from a camera with low MP, and upsize it to the number
of pixels of current high-MP cameras, and the noise becomes more
intense at the same resulting-pixel magnification.

I'm not sure what you mean by this. Upsampling does not make the
noise more intense at any fixed spatial frequency of the original
image.

Who is talking in that context?

I suggested that when you upsample a low-MP image to meet the higher-MP one, its real image noise, relative to the higher one, is now apparent, and no longer "hiding" by having a small displayed image size.

-- hide signature --

John

John Sheehy Forum Pro • Posts: 26,227
Re: To be more specific...

DaSigmaGuy wrote:

In thinking everyone was intelligent enough to know it, yes, but you
have proved that is'nt the case.

Can you laugh at yourself? That's a very useful skill, and you're going to need it, if and when you finally realize that your arguments here have been akin to a dog chasing its own tail.

Again, let me say that if we had 100MP monitors, and 1200 DPI dye-sub printers, we wouldn't be having this conversation. Your position would never have been held by any significant number of people.

-- hide signature --

John

DaSonyGuy Forum Pro • Posts: 12,352
Re: More pixels are not necessarily better...

Iliah Borg wrote:

The output image size is therefore determined purely by how many
pixels a sensor has

Totally wrong.

No, its perfectly correct.

The output image size is determined by the needs of
the photographer/publisher and by the acceptable quality criteria.

Of course you can upsize or downsize from the native image size but the fact remains that a 10mp sensor will produce a 10mp sized image and a 12mp sensor will produce a 12mp sized image etc.

Also, it is determined by the time and money spent on preparing the
image to print.

I was'nt talking about prints...Not everyone prints every pic they have and most will view them on computer monitor screens where the effect of pixel count on the output image size is most noticable.

I have 3 cameras with the same pixel counts. I can't print to the
same size from all of them, given the quality criteria and printing
method are established the same for all three.

Print count is irrelevent when printing. I can print a 4.6mp sized digital image at 2A0, without upsizing it first, and it will still look great...Showing pixel quality is far more important than pixel count, and with the effectiveness of new noise cancelling algorithms accepted its a fact that larger photosites offer better image quality and less noise than smaller photosites...If that was'nt true why would anyone buy DSLR's when they could save both bulk and money by buying high res digicams with tiny sensors!

-- hide signature --
 DaSonyGuy's gear list:DaSonyGuy's gear list
Sony a7C Sony FE 28-60mm F4-5.6 Samyang AF 24mm F1.8 FE Huawei P30 Pro +9 more
DaSonyGuy Forum Pro • Posts: 12,352
Re: To be more specific...

John Sheehy wrote:

ejmartin wrote:

John Sheehy wrote:

Take any noise from a camera with low MP, and upsize it to the number
of pixels of current high-MP cameras, and the noise becomes more
intense at the same resulting-pixel magnification.

I'm not sure what you mean by this. Upsampling does not make the
noise more intense at any fixed spatial frequency of the original
image.

Who is talking in that context?

I suggested that when you upsample a low-MP image to meet the
higher-MP one, its real image noise, relative to the higher one, is
now apparent, and no longer "hiding" by having a small displayed
image size.

John, have you even read his posts?...He has reached his opinion by comparing two different pics that he downloaded from some website that were'nt even exposed the same.

He then downsized the higher res 50D image to match the size of the lower res 40D image...That hides noise/reduces noise/gets rid of noise, whatever you want to call it, but the end result is totally irrelevant due to the flawed methodology used.

He then slags Phils 50D and 40D reviews off when they just happen to show his opinion to be incorrect.
--
DSG
--

-- hide signature --
 DaSonyGuy's gear list:DaSonyGuy's gear list
Sony a7C Sony FE 28-60mm F4-5.6 Samyang AF 24mm F1.8 FE Huawei P30 Pro +9 more
Iliah Borg Forum Pro • Posts: 29,167
Re: More pixels are not necessarily better...

The output image size is therefore determined purely by how many
pixels a sensor has

Totally wrong.

No, its perfectly correct.

Wait a minute...

Of course you can upsize or downsize from the native image size but
the fact remains that a 10mp sensor will produce a 10mp sized image
and a 12mp sensor will produce a 12mp sized image etc.

Here is where you are wrong. Sensors do not produce images. Image is something one can see. And since the image must be presented,...

I was'nt talking about prints

... you are dead wrong again in not talking about prints.

Not everyone prints every pic they
have and most will view them on computer monitor screens where the
effect of pixel count on the output image size is most noticable.

Display is just another output device, no different from a printer in many respects. Viewing the image, I'm not interested in looking at individual pixels. Only when I need to know how large I can print it I start looking closer.

Since you have obviously chosen to ignore that photography is about presenting the images at particular media through particular output process, not presenting pixels, please go ahead, I'll step aside and watch you losing your face further.

-- hide signature --
thw Veteran Member • Posts: 8,089
Amusing

DaSigmaGuy, I hope you realize you're arguing against Iliah Borg, EJ Martin and John Sheehy who are some of the most knowledgeable folks in these forums.

-- hide signature --

A View through my Lens
thw.smugmug.com

John Sheehy Forum Pro • Posts: 26,227
Re: More pixels are not necessarily better...

DaSigmaGuy wrote:

Of course you can upsize or downsize from the native image size but
the fact remains that a 10mp sensor will produce a 10mp sized image
and a 12mp sensor will produce a 12mp sized image etc.

While some people use the word "size" to refer to pixel counts, the usage is less than ideal. The number of pixels is the number of sampling points, not area dimensions. An image has no size, except whatever size you decide to display it at (and how close the viewer is to it).

The only reasonable way to address "image noise" and "image" resolution is to compare at the same size and distance, but that is almost impossible with our coarse monitor technology, without losing more potential resolution in the higher-MP images. So, people with your mindset engage in circular reasoning, and decide that the most that you should magnify a pixel is 100% pixel view, as if anything larger than that is somehow "forcing the image to a size it was never meant to be". That apparent limit, however, is self-serving, and protects the low-MP image from exposing how little resolution it has, and possibly how artificial its resolution is, as well as how much noise the image has, by forcing it to be smaller in comparisons.

We can't see our full resolution on monitors that I would allow in my budget (if any exist at all for 15 or 21 MP). So, in the meantime, the only fair way to compare image quality is to upsample the lower-MP image, preferably both, to the same image size, so that you are viewing the same percentage of the total image in two equal monitors or two windows on one monitor. You can step back and see how they look with less magnification.

Print count is irrelevent when printing. I can print a 4.6mp sized
digital image at 2A0, without upsizing it first, and it will still
look great...

Looks great to a Sigma Fan.

Showing pixel quality is far more important than pixel
count,

Nonsense, because that leaves you with an open equation. You need both, or you have absolutely ZERO information about image quality.

-- hide signature --

John

bobn2
bobn2 Forum Pro • Posts: 70,895
Re: Perhaps you can explain...

DaSigmaGuy wrote:

"Switching to our benchmark RAW converter, Adobe Camera RAW equalizes
image processing between the two cameras and allows us to get a much
better idea of the level of detail actually captured. As you can see
both cameras images look crisper and exhibit better detail but the
40D stills beats the newer model in terms of per pixel detail."

Bob does'nt think there is such a thing as "per pixel detail"...Are
you still prepared to go along with him now?

Well, in this thread I was disputing your use of the term 'pixel level image quality'. 'Pixel level detail' would be as clueless. 'Per-pixel x', as I said in my last post, makes a bit more sense, in that you can reference the quantity 'x' to pixel count, in the same way that the phrase '40 kilometres per hour' isn't necessaruly talking about any particular hour.

The quote continues:

[all this deleted, since EM, JPS and IB have covered it much better than I would have]

Having come back to this thread after being away to take some photographs (remember those) it's good to see you still in the fray, in addition to me, now blithely taking on the above, three of the most informed, expert and argumentative people on these forums, and still the quality of your arguments hasn't got any better. Brings in mind a story I heard...

A man takes his gun into the forest bear hunting. He sees a big grizzly, aims the gun and fires. When the smoke clears, he can't see the bear anymore, so he walks over to where it was to see if he's downed it - nothing.
Then he fells a tap on the shoulder, turns round, it's the bear!

The bear says: "Excuse me, my good man, I take a very dim view of being shot at, so now I'm going to tear off your limbs and eat you".
The hunter says: "Please, no. I've got a family to support, don't kill me".

The bear says: "Well, theres a bear's honour to think of. You tried to kill me and I've got to have retribution - it's the code. Tell you what though, if I can use your body to sate my depraved lusts, we'll call it quits."

So the man crawls home the next morning after a night of inconceivable depravity, with every part of his body aching. He's determined to get his own back, so he returns to the forest with an elephant gun. After a while, he spots the bear, aims and fires. There's a huge bang, and a recoil that nearly tears his arm off. When the smoke clears, no bear. Again he goes to see if he's got it. As he looks around he feels another tap on his shoulder.

"You again," says the bear, "this time, no choice, I'm going to rip your limbs off and eat you."
"No, think of my poor wife and children"

"Well...I had a pretty good time last night, so maybe I'll let you off again, just this once"

So the hunter staggers home the next morning. The bear's thought up all sorts of more extreme perversions and the hunter's ten times worse than the last night. Set on revenge, he returns to the forest with a bazooka.

Finds the bear, fires the bazooka. When the smoke clears, no bear. Goes to look for it, once again feels a tap on the shoulder.
"You're not really here for the hunting, are you?" says the bear.

Given that you seem to want to return to the forest time after time with only your first, and grossly inadequate firepower, I would have to conclude that you're not really here for the technical discussion.

-- hide signature --

Bob

ejmartin Veteran Member • Posts: 6,274
Re: To be more specific...

John Sheehy wrote:

ejmartin wrote:

John Sheehy wrote:

Take any noise from a camera with low MP, and upsize it to the number
of pixels of current high-MP cameras, and the noise becomes more
intense at the same resulting-pixel magnification.

I'm not sure what you mean by this. Upsampling does not make the
noise more intense at any fixed spatial frequency of the original
image.

Who is talking in that context?

Usually "the noise becomes more intense" means it becomes of higher amplitude.

I suggested that when you upsample a low-MP image to meet the
higher-MP one, its real image noise, relative to the higher one, is
now apparent, and no longer "hiding" by having a small displayed
image size.

I think perhaps you meant to say that the grain size is magnified, and so it is more visually apparent when the image is magnified, rather than the noise becoming "more intense". Just trying to preclude misinterpretation.

-- hide signature --
bobn2
bobn2 Forum Pro • Posts: 70,895
Re: SNR vs exposure

Graystar wrote:

bobn2 wrote:

I didn't say depends solely on sensor size and efficiency - that
would be silly. The floor of the DR is the read noise. Double that
between cameras (and it varies much more than that) and you have your
stop in DR.

So you're saying what...that the LX3 doesn't do any better in
highlights than the G10? It's all shadow detail?

Erm, no. DR gives a range of tone from 'highlight' to 'shadow'. The brightest 'highlight' you can render depends on the per-area FWC (Emil had a much better term for it, but I can't remember it now) and the darkest shadow depends on the noise. Exactly how the range of tonality of the original image is mapped into that space depends on the exposure index of the camera, and as we discover, under ISO standards the manufacturers can do that in any way they please.

-- hide signature --

Bob

Iliah Borg Forum Pro • Posts: 29,167
Arguing against photography

He is in fact arguing not against me, but against photographic use of the image. Not an isolated attempt to isolate an issue or a parameter to the point where it become totally irrelevant. Can't see forest for a tree.

Deep and widening gap between photography practice (and photography needs) and pixel peeping is very discouraging. Noise is one of the parameters that are easiest to measure, may be even the easiest one given current level of expertise of the peepers. Now that calibrationists can't even make a decent shot of a brick wall to compare resolution (and even less so - a decent shot of a decent resolution target) all we hear is noise. Noise, dynamic range, headroom in highlights is mostly all that we have here, and I'm yet to see any dissertation comparing current status of raw converters even in that regards, because needs for the output is totally ignored. Quality of midtones - they are not interested to find ways to describe and measure it. Plasticity of the image? - no. Same, colour is ignored. Same, artefacts, like moire, maze, are not evaluated. And even when they are discussing their pet peeve, noise, very little practical advice can be gathered from those boring discussions. Practical value of such discussions is dangerously close to a zero.

-- hide signature --
bobn2
bobn2 Forum Pro • Posts: 70,895
Re: Arguing against photography

Iliah Borg wrote:

He is in fact arguing not against me, but against photographic use of
the image. Not an isolated attempt to isolate an issue or a parameter
to the point where it become totally irrelevant. Can't see forest for
a tree.

Deep and widening gap between photography practice (and photography
needs) and pixel peeping is very discouraging. Noise is one of the
parameters that are easiest to measure, may be even the easiest one
given current level of expertise of the peepers. Now that
calibrationists can't even make a decent shot of a brick wall to
compare resolution (and even less so - a decent shot of a decent
resolution target) all we hear is noise. Noise, dynamic range,
headroom in highlights is mostly all that we have here, and I'm yet
to see any dissertation comparing current status of raw converters
even in that regards, because needs for the output is totally
ignored. Quality of midtones - they are not interested to find ways
to describe and measure it. Plasticity of the image? - no. Same,
colour is ignored. Same, artefacts, like moire, maze, are not
evaluated. And even when they are discussing their pet peeve, noise,
very little practical advice can be gathered from those boring
discussions. Practical value of such discussions is dangerously close
to a zero.

That is the nature of the review industry as a whole, cameras or any other commodity. If I read a review of a motor cycle, it will tell me how much power it puts out, how fast it accelerates, how long it takes to stop,everything that is easily quantified. It doesn't tell me anything about the riding experience, which is subjective and endlessly debatable. Anyone with any sense will use the quantitative measure to help draw up a shortlist, and then make the final selection by testing each themselves, and reputable dealers will let you test both cameras and motorcycles to help along your decision. People who don't have access to such a dealer or don't want to pay the inevitable price premium will substitute personal experience for the proffered 'expert' view of the reviewer.

Where the problem comes is when these 'experts' start proffering views which are in contradiction to the science and the evidence, for instance if motorcycle reviewers suggested that making the brake pads out of cheese improved 'stopping quality', even though it was obvious that measured stopping distance was far worse. This happened in high end audio, where the reviewers started suggesting that the measurements were meaningless, and only their ears could determine good sound quality and I would contend that it is beginning to happen in photography also, with DPR being amongst the prime culprits (although their approach is more subtle, they selectively present bogus measurements to support their subjective opinions - leading the naive to believe that they are just choosing between alternative sciences).

So, while I fully support your contention that there is much, much, more to it than just the theory and easily quantified parameters of image quality, that theory and those parameters do form the baseline from which more subjective evaluation must build. In particular, if the subjective evaluation appears to tell you something different from the quantitative evaluation, you need to be fully cognisant of its subjectivity before extending it to a sweeping generalisation.

-- hide signature --

Bob

John Sheehy Forum Pro • Posts: 26,227
Re: To be more specific...

DaSigmaGuy wrote:

John, have you even read his posts?...He has reached his opinion by
comparing two different pics that he downloaded from some website
that were'nt even exposed the same.
He then downsized the higher res 50D image to match the size of the
lower res 40D image...That hides noise/reduces noise/gets rid of
noise, whatever you want to call it, but the end result is totally
irrelevant due to the flawed methodology used.
He then slags Phils 50D and 40D reviews off when they just happen to
show his opinion to be incorrect.

The general principles are already known without his particular example. Had his posts never existed, the issues would still be the same.

My 30D, taking the same shot with the same shutter speed, same f-stop, same tripod-mounted lens, gives the same noise at the same image size as my 50D, whether I upsample both downsample both, etc. The only difference is resolution, when it isn't downsampled away in the 50D.

-- hide signature --

John

John Sheehy Forum Pro • Posts: 26,227
Re: To be more specific...

ejmartin wrote:

I think perhaps you meant to say that the grain size is magnified,
and so it is more visually apparent when the image is magnified,
rather than the noise becoming "more intense". Just trying to
preclude misinterpretation.

Fair enough. Poor choice of word.

-- hide signature --

John

bobn2
bobn2 Forum Pro • Posts: 70,895
Re: To be more specific...

John Sheehy wrote:

DaSigmaGuy wrote:

He then slags Phils 50D and 40D reviews off when they just happen to
show his opinion to be incorrect.

The general principles are already known without his particular
example. Had his posts never existed, the issues would still be the
same.

and more to the point, Phil's reviews don't show Emil's posts to be incorrect, Emil's posts show Phil's reviews to be incorrect.

-- hide signature --

Bob

Ralf Ronander
Ralf Ronander Contributing Member • Posts: 797
Re: To be more specific...

There you go, arguing as if I had argued against your conclusions, while I specifically said I do not argue against them. Your conclusions are well known around here, no need to repeat them.

Now you say you evaluated noise in a specific area of the images, but you didn´t say that in the post to which I responded. However, you said noise spectrum was equal for both images. Not easy to know you meant just certain parts of said images.

The relevance of the placement of certain items in the scene is with regard to resolution; I´m sorry for having missed that you with reference to resolution had other images in mind.

Laurentiu Todie
Laurentiu Todie Senior Member • Posts: 2,567
Re: More pixels are not necessarily better... but can be

a one square inch crop at 72dpi off a detailed scene (textile fiber?) could settle it : )

Keyboard shortcuts:
FForum MMy threads