More megapixels, better photos: Fact or fiction?

Started Feb 7, 2008 | Discussions
MjDon Contributing Member • Posts: 735
Fiction since most people take terrible photos with any camera. nt

nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn

PBasepeterb Senior Member • Posts: 1,439
Re: only LARGE prints matter

JoePhoto wrote:

BLawson wrote:

PBasepeterb wrote:

More MP = potentially better prints
but not necessarily better photos.

...how you can have a "better print" that is not a better
photograph? Output is the objective of photography; isn't an
unprinted photograph an oyxymoron, like unplayed music?

I may not understand what you mean .... but I will try a response.

More mega-pixels is ONLY useful on "prints" .... and more
specifically LARGE-prints.

The problem is that so very many people simply look at an image, and
make their subjective decision based on only that.

Yet a resolution as low as 72dpi is sufficient for monitor display,
whereas prints can utilize up to 300dpi.

So any discussion referring to necessary mega-pixels are meaningless
for monitor ... or even up to 8x10's; it is 20x30, 24x36, and 30x40
where it makes a difference.

That's basically what I was trying to get at Joe...thanks for expanding
on the thought.

Brian, as far as unprinted "photo" being an oxymoron...I see your point.
I suppose one needs to make a distinction between a "photograph" which
is a printed object and a "digital image" which is something I can view on
my monitor and transmit electronically.

More MPs can benefit the former but not the later so much (for the reasons
Joe points out) .

-- hide signature --

Keep your lens clean and your mind open.

http://www.pbase.com/peterb/

bkj Senior Member • Posts: 1,339
Re: Get a Mamiya if you need more megapixels.

Manufacturers should concentrate their R&D and marketing efforts on the quality of pixel, not the quantity of pixel.

For example, they should concentrate on designs that can truly reduce noise, not a processor that smudges the details to cover up the noise.

If people want more megapixel, they should switch to a larger sensor format instead of jamming more megapixel into a tiny sensor.

For example, there're landscape photographers that always want more megapixels. Instead of getting a DSLR, these folks should get a medium format camera with a digital back. The Mamiya 645 package is only $9999, which includes the following items:

Mamiya ZD 645 AFD II Camera
Mamiya 22MP Digital Back (sensor made by Dalsa)
Mamiya AF 80mm f/2.8 lens

Add the Mamiya AF 28mm f/4.5 (pretty hefty at $5K though) to the above package, they now have a 22MP medium format package that's good for both landscapes and portraits.

$9999 may sound expensive, but considering the Canon 1DIII costs $7999, getting a digital medium format package for $9999 with one lens doesn't sound too bad.

Cheers

-- hide signature --

Fujifilm S5 Pro, S3 Pro
Tokina AF 12-24/4.0
Nikon AF-S 18-200/3.5-5.6G VR
Tamron AF 28-75/2.8
Nikon AF 50/1.4D

http://www.flickr.com/photos/autoclavable/

John Sheehy Forum Pro • Posts: 21,729
Re: An image of course

Chato wrote:

John Sheehy wrote:

Chato wrote:

Nice picture?

The IQ of that picture is horrible by current 10MP P&S standards.

is judged by the way it prints. There's nothing "plastic" in this
image, which is what you seem to be looking for.

It's totally plastic. Sharpening halos at a fairly wide radius, textures that look like melted wax.

The composition is mediocre. Is this some kind of a joke? It looked
better when it was small - it has much to hide.

The composition of if course Great. But that's a matter of taste.

You seem to take pride in mediocrity. Your most famous comment on
Foveon sensors

My most famous comment? You mean the comment I made that stood out in your mind. Don't drag the rest of the world into this.

if that they are interpolated up, then the viewer will
see a "mosiac." A comment that you still stand by in the face of all
the obvious evidence to the contrary.

The evidence is quite clear to me; images that are aliased upon capture show it more, the more you enlarge, regardless of the upsampling algorithm. The level of aliasing possible in a sigma digital camera is far stronger than what is possible in a camera with an AA filter, and I can see right away that the level of aliasing goes beyond the norm; the effects of sampling less than 100% of the jurisdiction of each pixel is obvious in areas of sharp focus. Aliased capture, especially of low MP count, has no ability to properly locate edges. An AA-filtered camera can place the center of light in points and edges at any analog position between pixels; an aliased imager can not, especially when the 2-D spatial pixel count is low.

On this thread you argue that more MP's equals better IQ, no matter
what the sensor size.

That's a very poor paraphrase of what I might have said. I have said that in a given sensor size, there is no increase in image shot noise when the pixel density is increased other than what might be lost to quantum efficiency (which is obviously still far from where technology stands now; some 2 micron P&S photosites have some of the best quantum efficiencies at this point in time). And also, if the product of the read noise at the pixel level, and the dot pitch does not increase, then the image read noise does not increase. The product, in general, has gone down or remained the same for most lines of cameras with increasing pixel density.

Once again stating this

I'm not sure you understand what it is that I state. You seem to have a very inelastic mind, incapable of wrapping itself around any paradigm other than the very first one it clinged onto. If I gave a test on what I stated, I'm sure you would fail.

despite all visual
evidence to the contrary. Well, I can't argue with you from a
technical aspect - All I can do is see what you are referring to.
Good IQ becomes a Mosaic, bad IQ becomes magnifiscent.

All a bunch of high-order abstractions and distortions of what I have said.

-- hide signature --

John

John Sheehy Forum Pro • Posts: 21,729
Re: An image of course

Chato wrote:

Could there be a reason why all these high MP camera come with heavy
NR, built in?

All P&S cameras have had heavy NR, all along. You didn't notice?

It's just become a little bit more frantic and ridiculous lately, and more noticable as people "peep" to see what their extra pixels are getting for them.

I have enumerated some of the reasons for the NR race many times in other threads, and touched on them here.

1) People look at images at 100% on the monitor, and have no idea of the significance of that.

2) Even if someone views the image scaled to fit a fixed desktop size or window size, the downsizing algorithms used to view them often weight input pixels unequally in the output, and maintain original pixel-level noise (which should actually reduce at a certain rate in downsampling). These are usually the methods that give sharp results. The downsizing algorithm used in the display even in $600 Photoshop is a joke from the days when memory and CPU power were just a fraction of what systems have now. Make an image that is a checker-board with one-pixel checks and resize it in Photoshop's "Navigator" with the slider, to see how bad it really is. Where you should see grey or mild modulations of grey, you see black-and-white checks of arbitrary size and doily patterns. Ideally, we'd be upsampling our high-MP images to fill the higher-MP screen; not downsizing them.

3) People allow the software and print drivers to do the resizing at print time, instead of doing it themselves. Print drivers unnecessarily resample images to an intermediate PPI before being translated to their dot arrays; a very bad policy if you ask me; the translation should be direct, as accurate local color per output "pixel" becomes less necessary at high resolutions.

4) There are gamma issues involved in viewing raw readout noise. The sRGB standard is to use gamma-correction above a certain level, and linear below, but the read noise in the deepest shadows actually requires a gamma adjustment in the opposite direction of the one used in the higher tones, and many people's monitors have these lower ranges coming up too bright to begin with.

5) Lots of people just remember fond things from the past, and everything bad seems to be happening "these days"; a purely psychological filter.

Remember the good old days?

Nothing but psychology can explain your fondness for the IQ of the image you embedded (bridge and bottom-lit clouds).

No of course not, it's the public that demands plastic
that drives the question. The manufacturers are merely meeting the
publics demands for NR - Other wise the images would be as awesome as
my little Two MP Olympus put out.

There's nothing awesome about that image, IMO; especially IQ-wise. I feel like I'm looking at pixels and artifacts. You could have at least done something with the saturation and contrast, to make it look like it does in person.

There's nothing wrong with the MFRs meeting the public demand, but it should be optional, and the cameras should have some fairly literal conversion as an option, too. Most do not. Their "none, low, and high" NRs are really "high, higher, and higher yet".

-- hide signature --

John

John Sheehy Forum Pro • Posts: 21,729
Re: only LARGE prints matter

PBasepeterb wrote:

More MPs can benefit the former but not the later so much (for the
reasons
Joe points out) .

Someday you might be looking at your images on a monitor that can display every pixel, though.

Still, even in the current day, with our 100 PPI monitors, a higher-resolution original allows greater flexibility in resampling with minimal artifacts.

-- hide signature --

John

Ehrik Veteran Member • Posts: 8,014
Re: I agree it depends

Eric Fossum wrote:

There is still a factor of 2x in QE to be gained, at least,

With a Bayer sensor? Or by replacing the colour filters with something
that doesn't waste photons?

and a factor of 5x in read noise,

So a "clean" photon counter is not possible?

and a factor of 5x in full well

I don't see why there is a theoretical limit. If the sensor is built more in 3D
and the electrons are led deep down? Or if the well is emptied and refilled
several times during exposure and a counter keeps track. Of course in this
case there is a full well conceptually even if it's not limiting.

Sorry for being so curious.

before I
think we will start to hit real limits in incremental improvements.
Just a guess.

Just my two oere
Erik from Sweden

luvnozzy Regular Member • Posts: 415
Re: More MP

i didn't realize they had color photography back then. who passed this image on to you?

PBasepeterb wrote:

PBasepeterb Senior Member • Posts: 1,439
Re: More MP

I just hopped into my Wayback Machine.... ;> )
--
Keep your lens clean and your mind open.

http://www.pbase.com/peterb/

John Sheehy Forum Pro • Posts: 21,729
Re: I agree it depends

Ehrik wrote:

So a "clean" photon counter is not possible?

It seems to be very difficult for large photosites. A sensor that could just count photons with high quantum efficiency would be miraculous, at least as far as shadows are concerned. The cleanest "counters" are "photo avalanche diodes", AFAIK, as they have normal read noise in electrons, but each photon causes multiple electron charges, so they have low read noise in photons.

and a factor of 5x in full well

I don't see why there is a theoretical limit. If the sensor is built
more in 3D
and the electrons are led deep down?

Since this is the most obvious solution (but only for having lower ISOs, of course), and it hasn't had much improvement, I would assume that there are obstacles in this approach.

Or if the well is emptied and
refilled
several times during exposure and a counter keeps track. Of course in
this
case there is a full well conceptually even if it's not limiting.

Refilling might cause glitches in counting, make more noise in the sensor, require too much processing, etc. It is an easy concept, but the implementation may be difficult.
--
John

luvnozzy Regular Member • Posts: 415
we should colaborate...

i've been busy in my garage for a couple of years now.

don't you just love h.g. wells? what a visionary.

PBasepeterb wrote:

I just hopped into my Wayback Machine.... ;> )
--
Keep your lens clean and your mind open.

http://www.pbase.com/peterb/

Chato
Chato Forum Pro • Posts: 46,792
Re: An image of course

Jay Turberville wrote:

The IQ of that picture is horrible by current 10MP P&S standards.

is judged by the way it prints. There's nothing "plastic" in this
image, which is what you seem to be looking for.

The image is mostly near black with very little area to judge detail.

Yes, of course - Obvious is it not?

But regardless, the comparisons you've made are poor since the
subject changes significantly and the images are too small to
appropriately judge.

That's the full sizes image. Somewhere I have the original. This was from my print folder. And I down sized it from 5600(?) pixels, and then saved it at Photoshops 50 percent compression. It prints much more like the smaller image I posted.

On this thread you argue that more MP's equals better IQ, no matter
what the sensor size.

Actually, he never made that argument. Nor did I.

-- hide signature --

His argument is a cliché, All things being equal, the more MP the better. Whoa, how original! (now if only we could get rid of those technical problems that manufacturers seem to have a problem with)

Dave

Chato
Chato Forum Pro • Posts: 46,792
Re: An image of course

John Sheehy wrote:

3) People allow the software and print drivers to do the resizing at
print time, instead of doing it themselves. Print drivers
unnecessarily resample images to an intermediate PPI before being
translated to their dot arrays; a very bad policy if you ask me; the
translation should be direct, as accurate local color per output
"pixel" becomes less necessary at high resolutions.

I use a form of stochastic printing. Software converts the image to individual dots, (in other words a "fixed" image) which in turn get printed one for one...

4) There are gamma issues involved in viewing raw readout noise. The
sRGB standard is to use gamma-correction above a certain level, and
linear below, but the read noise in the deepest shadows actually
requires a gamma adjustment in the opposite direction of the one used
in the higher tones, and many people's monitors have these lower
ranges coming up too bright to begin with.

And? We all know that you don't print what you "see" on the monitor. And there are a hell of a lot more reasons for that then your simple summary.

5) Lots of people just remember fond things from the past, and
everything bad seems to be happening "these days"; a purely
psychological filter.

Remember the good old days?

Nothing but psychology can explain your fondness for the IQ of the
image you embedded (bridge and bottom-lit clouds).

Nothing but the way it prints AND the way it impacts views. That would seem simple enough to any unbised person? I see that kind of light maybe once every five years. The first example does it better justice than the larger - But it PRINTS like the first example. You don't like it? Good for you. I never object to someones personal opinion.

No of course not, it's the public that demands plastic
that drives the question. The manufacturers are merely meeting the
publics demands for NR - Other wise the images would be as awesome as
my little Two MP Olympus put out.

There's nothing wrong with the MFRs meeting the public demand, but it
should be optional, and the cameras should have some fairly literal
conversion as an option, too. Most do not. Their "none, low, and
high" NRs are really "high, higher, and higher yet".

When I call an image I see "plastic" it is because there are no visible details. Objects of saturated color, forms that you can tell are a bird. Indeed it's easy to recognise Uncle Harry or the family dog. They are plastic because they PRINT plastic.

John

Dave

Chato
Chato Forum Pro • Posts: 46,792
Re: Well Chato ... here is a good comparison...

Jay Turberville wrote:

Hey, why argue with me? Play my game yourself -

Now let's play my game. In my game I take basically the same picture
at nearly the same time with three different cameras. In this case,
a Coolpix 5000 with a 5Mp 2/3" sensor, a Coolpix 8400 with an 8Mp
2/3" sensor and an E-330 with a 7.5Mp 4/3" sensor. I used
approximately f/4 on the two 2/3" cameras. The E-330 image was shot
at f/8. All were shot at the camera's lowest ISO - hence the highest
quality. That's ISO 50 for the CP8400 and ISO 100 for the other two.
All images were shot with raw, developed in dcraw as 48 bit
non-gammma adjusted TIFFs and color adjusted in Photoshop 7. No
noise reduction was applied at any step. Sharpening was done to the
luminance channel in Lab color space and 48 bit color depth. The 5Mp
and 7.5Mp images were uprezzed to 8Mp.

I'm only posting two images. One is a small picture of the full
scene. The other shows an image crop at 100% pixels. I'll provide
links to other crops as well as links to full rez final copies. But
this way, this page won't take forever to load.

BTW, I could have pulled more detail out of the CP5000 image by
sharpening a bit more, but that would have further increased the
noise and exaggerated some of the aliasing

The summary here is that the CP5000 has less detail and more noise
than the CP8400 - the camera with smaller pixels. The E-330 has the
least amount of noise, but also is a bit less sharp. A better lens
might help the E-330 a bit or it might just be that the E-330 has a
more agressive AA filter. I don't know.

But anyway, this demonstrates my point - and the camera makers' point
rather well. Increasing pixel count on the same sized sensor can
improve overall image quality. In this case, it improves resolution
and noise - though it probably gives up a little on the DR front. Of
course, improvement may not always be the case. Another comparison
with different cameras might give a different result. The lesson
here is that you can't just assume. You need to actually do some
checking.

Now, from the standpoint of making an 8x10 or 11x14 print, all of
these cameras can give very nice output. The practical differences
really aren't that large. But overall image quality improved with
the CP8400 and its smaller pixels. It didn't get worse. In good
light, I don't hesitate to use the CP8400 if the situation warrants
it. Its images don't look plastic (take a look at the 4 comparison
images) and though it has a tad more noise than my DSLRs, it isn't
enough to hardly ever matter.

The image order is:

Coolpix 5000................Coolpix 8400 ....................E-330

These are around 450kb.
http://www.jayandwanda.com/dpreview/E-330_v_CP8400/CP5K_v_CP8400_v_E-330_3.jpg
http://www.jayandwanda.com/dpreview/E-330_v_CP8400/CP5K_v_CP8400_v_E-330_4.jpg

These files are 6-7Mb and my server runs off of my DSL connection.
So don't expect fast uploads.

http://www.jayandwanda.com/dpreview/E-330_v_CP8400/TT103877_Mod_.jpg
http://www.jayandwanda.com/dpreview/E-330_v_CP8400/dscn1420_Mod.jpg
http://www.jayandwanda.com/dpreview/E-330_v_CP8400/dscn1887_Mod.jpg

http://www.jayandwanda.com/dpreview/E-330_v_CP8400/TT103877_Mod_.jpg

I'll check out your site this weekend - But which of the three above "prints" better?

Dave

Eric Fossum
Eric Fossum Senior Member • Posts: 1,380
Re: I agree it depends

Ehrik wrote:

Eric Fossum wrote:

There is still a factor of 2x in QE to be gained, at least,

With a Bayer sensor? Or by replacing the colour filters with something
that doesn't waste photons?

Well, between silicon device engineering, optics, and perhaps filter engineering. I am still thinking in the context of RGB CFAs.

and a factor of 5x in read noise,

So a "clean" photon counter is not possible?

Again, I am talking about incremental improvements. Buried channel source-follower or p-channel SF are possible ways to help.

and a factor of 5x in full well

Again, incremental improvements. Also, swing on the SF output and signal chain, and conversion gain are also related to full well. All 3 pieces have to match. No sense having a huge full well if it saturates the output amp, right? And conversion gain (uV/e-) needs to be high to reduce read noise.

I don't see why there is a theoretical limit. If the sensor is built
more in 3D
and the electrons are led deep down? Or if the well is emptied and
refilled
several times during exposure and a counter keeps track. Of course in
this
case there is a full well conceptually even if it's not limiting.

Well, 1.4 or less micron pixel pitch does not leave a lot of room for other stuff in the pixel. If you are asking about more dynamic range that is a different story and related to other factors besides pixel design.

Sorry for being so curious.

Hardly a problem!

 Eric Fossum's gear list:Eric Fossum's gear list
Nikon Coolpix P900 Sony RX100 II
John Sheehy Forum Pro • Posts: 21,729
Re: An image of course

Chato wrote:

His argument is a cliché, All things being equal, the more MP the
better. Whoa, how original! (now if only we could get rid of those
technical problems that manufacturers seem to have a problem with)

No, Dave; you don't even seem to understand what my argument is, or you're trying to make me look stupid by creating the illusion that I'm rambling on about something that everyone already knows. No one, including myself, has disputed the fact that more pixels of the same quality each, is always better.

My argument is that you can maintain the same image noise, even while increasing pixel density, with the noise per pixel actually increasing to some degree. This happens automatically with shot noise, if you maintain the same QE, because that's just how the math works out. 4 pixels in the space of one means 1/4 as many photons each, and hence double the shot noise (relative to saturation). 4 pixels combined divides the shot noise by 2, and you're back to the original shot noise. Another way to look at it is that For read noise, things can actually improve with a higher pixel density, as real world read noise does not have to increase as fast as shot noise does at the pixel level, when you increase pixel density. If read noise increases by the ratio of the old pixel pitch to the new pixel pitch, image read noise will remain the same. If the pixel read noise of the higher-density pixels remains the same, then the image read noise will decrease. If the read noise per pixel actually decreases (as it seems to do in the Pentax K20D) then image read noise reduces even more so.

-- hide signature --

John

John Sheehy Forum Pro • Posts: 21,729
Re: Well Chato ... here is a good comparison...

Chato wrote:

I'll check out your site this weekend - But which of the three above
"prints" better?

The middle and the right might be difficult to choose between, because one has less noise, but also has less detail. The left one is definitely out, though. It is worthless if you have the other two.

-- hide signature --

John

Jay Turberville Forum Pro • Posts: 12,917
Re: Well Chato ... here is a good comparison...

I'll check out your site this weekend - But which of the three above
"prints" better?

The middle and the right might be difficult to choose between,
because one has less noise, but also has less detail.

The left one (Coolpix 5000) also has noticable aliasing artifacts in the power lines. So, compared to the other two, it is quite clearly inferior. That said, on my lapto that has a 13"x8" (1280x768) screen, 100% pixels is about 100 ppi and I am looking at the equivalent of a 35" x 25" print from less than a couple feet away. Even though it suffers compared to the E-330 and Coolpix 8400, the Coolpix 5000 still makes a very nice 11x14 print.

The left one
is definitely out, though. It is worthless if you have the other two.

Right. In comparison it is inferior.

I won't be actually making prints. The image isn't that interesting.

I've been mulling this over today and right now my suspicion is that you can probably continue to improve overall image quality by reducing pixel pitch (assuming you use the same or better technology) up to the point where the sensor clearly outresolves the camera lens. After that, further pixel pitch increases probably don't help much. Also, once the sensor clearly outresolves the lens, you can drop the use of the AA filter, improving overall sharpness a bit - which is probably the last little bonus you get for having more pixels. People seem to forget that putting a filter to purposely blur the image projected by a lens is a significant compromise.

Does anybody want me to compare my Coolpix 995 with a 3Mp 1/1.8" sensor to my Olympus C7070 which has a 7Mp 1/1.8" sensor? Or are we all pretty convinced that the C7070 will give superior results?

-- hide signature --
Jay Turberville Forum Pro • Posts: 12,917
Re: I agree it depends

Well, 1.4 or less micron pixel pitch does not leave a lot of room for
other stuff in the pixel. If you are asking about more dynamic range
that is a different story and related to other factors besides pixel
design.

Perhaps, but they've been finding more room as they've moved to 1.75 micron pixels pitches. According to Sony spec sheets, they have kept the photodiode size constant or slightly increased its size as they went from 2.2 micron to 1.86 micron sensor designs. And they have been reducing pixel pitch while increasing the relative photodiode size for quite some time. I don't have anything spec sheets more current than the 1.86 design, but 1.86 microns gets us about 11Mp on a 1/1.8" sensor - something that people seems inclined to want to complain about. But why? The per pixel performance should be pretty similar to their 7Mp sensor - and you get about 50% more pixels!!! How could image quality not improve?

The 12Mp 1/1.7" sensor from Sony uses 1.75 micron pixels. These pixels increased the aperture ratio by 9% while decreasing the pixel pitch. People have this impression that while the pixel pitch decreases that the size of the light sensitive photodiode also must decrease. Well it isn't necessarily so.

-- hide signature --
Zoltan2 Senior Member • Posts: 2,991
Ken Rockwell

Read his article: http://kenrockwell.com/tech/mpmyth.htm
--
pbase Supporter

Keyboard shortcuts:
FForum MMy threads