-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
Kodad Samples - Magnification fools us!
- Thread starter Geir Ove
- Start date
I fail to see how posting a link to another message that cites the first message in this thread helps to advance the discussion at all.
Matt Difanis
Forum Enthusiast
If this is such simple math, then how do you figure that a 14 megapixel image is magnified by a factor of 16 on your monitor, whereas a 6 megapixel image is magnified three times? 14 is 2.33 times larger than 6, yet your figures suggest some other ratio. You didn't do accounting at Enron, did you? (Just kidding in a good natured sort of way.)
The bottom line is that an image sensor ought to capture real, meaningful data for each and every effective pixel of that sensor. On a computer monitor, viewing the image at 100% or larger is the only way to examine an image on a pixel-by-pixel basis. Whether the image sensor that created the image is a 14 megapixel, a 2 megapixel, or a 6 megapixel, when displaying a non-upsampled image at 100% size on a computer monitor, the image should look clear and as detailed as the monitor is capable of displaying, regardless of how large a chunk of the total image is on your screen at any one moment.
Just the thoughts of a guy who's in law school to avoid math. Now you've made my head hurt.
Matt
Hello Matt,
The numbers you referred to where approximate figures, not a computation.
The link to the explanation is here:
http://www.objective.no/geirove/sensor_screen_mismatch.htm
Here I have a more accurate computation:
quote:
Thus D60 picture is 3 times as wide as the Display when displayed at 100% size (1:1).
Thus 14n picture is 4,43 times as wide as the Display when displayed at 100% size (1:1). > >
To see how many screens it fills, you have to do the same in the Vertical direction, and multiply the numbers:
Assume Vertical screen res = 768.
D60 vertical res = 2048 => 2048 / 768 = 2.66
14 N vertical res = 3024 => 3024 / 768 = 3.93
The perceived sizes of the 14n and D60 pictures viewed at 100% at a Display with 1024*768 reslution then becomes:
D60: 3 * 2.66 = 7.98 screens
14n: 4.43 * 3.93 = 17.41 screens
The ratio becomes : 17.41 / 7.98 = 2.18
This makes sense, since the ratio of the resolutions is approx:
13.9 / 6.29 ~ 2.2
Not fuzzy at all
(Enron: No, we don't do bookkeeping or math like this in Norway )
Geir Ove
The numbers you referred to where approximate figures, not a computation.
The link to the explanation is here:
http://www.objective.no/geirove/sensor_screen_mismatch.htm
Here I have a more accurate computation:
quote:
Thus D60 picture is 3 times as wide as the Display when displayed at 100% size (1:1).
Thus 14n picture is 4,43 times as wide as the Display when displayed at 100% size (1:1). > >
To see how many screens it fills, you have to do the same in the Vertical direction, and multiply the numbers:
Assume Vertical screen res = 768.
D60 vertical res = 2048 => 2048 / 768 = 2.66
14 N vertical res = 3024 => 3024 / 768 = 3.93
The perceived sizes of the 14n and D60 pictures viewed at 100% at a Display with 1024*768 reslution then becomes:
D60: 3 * 2.66 = 7.98 screens
14n: 4.43 * 3.93 = 17.41 screens
The ratio becomes : 17.41 / 7.98 = 2.18
This makes sense, since the ratio of the resolutions is approx:
13.9 / 6.29 ~ 2.2
Not fuzzy at all
(Enron: No, we don't do bookkeeping or math like this in Norway
)Geir Ove
pcockerell
Forum Enthusiast
I confess I didn't read every word (I find the sheer volume of your replies overwhelming, to be honest), but two things need commenting on:
Anyway, moire patterns aren't only not specifically Bayer artifacts, they're not even specifically a digital imaging phenomenom. It's an interference pattern caused by the "beating" (to use a signal processign term) of two patterns of similar form in close alignment. This cute applet illustrates it:
http://math.hws.edu/xJava/other/Moire1.html
Note that the patterns you see would be present even if the lines were being rendered at some infinite resolution; it's your eyes that are producing the patterns in this case, not the pixels used to render them. Note also that the lines in this example are monochomatic. No Bayer masking here. (There are numerous toys with patterned transparent sheets with circular and other patterns printed on them in distinctly non-Bayered black that explot the moire effect.)
But yes, moire is also used in digital imaging to refer to the interference pattern caused when the image detail is close in frequency to the sampling grid and aligned at small angles to it (and there's no AA filter to remove those high frequencies). But it's just as present on a monochrome sensor as it is on a Bayer one. Or, as your link shows, a Foveon X3. It's true that an extra dimension of moire patterns on Bayer sensors is that they exhibit color fringes (because you've effectively got three separate sampling grids of three different colors and two different frequencies), but I wasn't referring to that in my original comment; I was just referring to moire fringing in general.
And...
As for whether the 14n's noise reduction artifacts are visible in printing or not, we'll just have to agree to differ. I have three pictures in front of me printed on a not very high end printer at 300dpi and they're clearly visible at 50% and 100% NR. I can only assume that anyone using a 14n for a variety of different uses, from fine art photography to double-page advertising spreads in high-end glossy magazines would find the same thing, whether they can see them on a monitor preview or not.
Of course, as we now know, no-one in their right mind would apply the 50% level of noise filtering (and hopefully it wouldn't be necessary), so the point is moot. But that doesn't make Geir's original assertion that the lousy quality was due to the "unfair" viewing of the images at 1:1 any more valid, and never will.
Cheers,
Pete
--
http://www.pbase.com/pcockerell
http://www.peter-cockerell.net:8080/
Well, it's always nice to have an article that actually bolsters my point. Two observations about that article are that (a) it doesn't contain the word "Bayer", and (b) it uses the distinctly non-Bayer Sigma SD9 to illustrate the concept or moire patterns. Hmmm.
Anyway, moire patterns aren't only not specifically Bayer artifacts, they're not even specifically a digital imaging phenomenom. It's an interference pattern caused by the "beating" (to use a signal processign term) of two patterns of similar form in close alignment. This cute applet illustrates it:
http://math.hws.edu/xJava/other/Moire1.html
Note that the patterns you see would be present even if the lines were being rendered at some infinite resolution; it's your eyes that are producing the patterns in this case, not the pixels used to render them. Note also that the lines in this example are monochomatic. No Bayer masking here. (There are numerous toys with patterned transparent sheets with circular and other patterns printed on them in distinctly non-Bayered black that explot the moire effect.)
But yes, moire is also used in digital imaging to refer to the interference pattern caused when the image detail is close in frequency to the sampling grid and aligned at small angles to it (and there's no AA filter to remove those high frequencies). But it's just as present on a monochrome sensor as it is on a Bayer one. Or, as your link shows, a Foveon X3. It's true that an extra dimension of moire patterns on Bayer sensors is that they exhibit color fringes (because you've effectively got three separate sampling grids of three different colors and two different frequencies), but I wasn't referring to that in my original comment; I was just referring to moire fringing in general.
And...
Eh? This is a ratio of > 1/125, or if my monitor is 96dpi, the print device you're referring to has at least 12,000 dpi resolution. I think you've misplaced a decimal position or two here.
As for whether the 14n's noise reduction artifacts are visible in printing or not, we'll just have to agree to differ. I have three pictures in front of me printed on a not very high end printer at 300dpi and they're clearly visible at 50% and 100% NR. I can only assume that anyone using a 14n for a variety of different uses, from fine art photography to double-page advertising spreads in high-end glossy magazines would find the same thing, whether they can see them on a monitor preview or not.
Of course, as we now know, no-one in their right mind would apply the 50% level of noise filtering (and hopefully it wouldn't be necessary), so the point is moot. But that doesn't make Geir's original assertion that the lousy quality was due to the "unfair" viewing of the images at 1:1 any more valid, and never will.
Cheers,
Pete
--
http://www.pbase.com/pcockerell
http://www.peter-cockerell.net:8080/
Rylee Isitt
Senior Member
I'm "from" the Canon forum, but really I don't live there, so don't let that scare you away. I'm in my second year of digital arts at Ai CDIS, so I can somewhat understand what you're saying.
Your theory holds true on the screen only, loosely. But a 100% view from, say, a 1Ds (11 MP), looks far superior to the samples of Kodak's camera.
As someone else has mentioned, if we just wanted to view stuff on screen, there would be no point to have increased resolution beyond about 1024*768, or whatever most people use on their monitors. You are not increasing the resolution of your monitor when you downsize your image - you are still viewing the same resolution.
Remember, there are two seperate "resolutions": the physical size (in pixels) of your image, and the resolution of the medium that's displaying it (eg, 1024x768 on many monitors).
When you "shrink" the picture down to fit on the whole screen, you are actually NOT making the pixels smaller. You're actually loosing detail - lots of it. But since the detail you are loosing is not there in the first place in the case of these sample images, the image looks better since it's being viewed at a size where the details look normal, instead of just an enlarged 2MP image.
What you are seeing on the monitor with a picture fit to the screen can convey no more detail than a 100% view can since the monitor is set at a fixed resolution of roughly 72dpi (changes from one screen to the next).
A camera should be able to produce pictures which look as good at full size then they are fit to the screen.
So, yes, your massively MP camera is producing a fine 2MP image, but in reality, a 2MP image gets you nowhere past viewing it on a monitor or using it on the web.
Printing is a critical part of photography, and print keeps the detail of the image. You can print at any DPI you want, limited only to your tools and/or budget. You can print at 600dpi and keep every single detail - including every single artifact. Sure, when you stand back and look at the print as a whole the problems may not be noticeable, but inspect it with a magnifier, and they're just as bad as they were in the samples we've seen.
Want to print posters with that camera? Sure, you'd only be able to print them at a low DPI, but those problems are going to look terrible.
Want to print 8x10 crops? The problems are still there.
Printing smaller? Well, then you're probably safe.
the point is, for an 14MP camera, it's acting like it's 2MP. That's not good. The 1Ds sure doesn't act like a 2Mp or even a 6MP camera. It acts like you'd expect - even at 100% views on screen, no artifacts... nothing but detail.
However, I suspect the samples you have seen of the 14N are worse than what the camera can output. It looks like they've had noise problems, and using noise reduction has killed any fine detail the image had except in the high contrast areas. Hopefully the delays are going to result in those noise problems being fixed.
I hope the 14N turns out to be a jewel. Canon could use some harsh competition - so could everyone. dSLR prices are just outragous right now, at least for a guy without any cash.
Good luck you! Now stop worrying about megapixels an go take some pictures!
Your theory holds true on the screen only, loosely. But a 100% view from, say, a 1Ds (11 MP), looks far superior to the samples of Kodak's camera.
As someone else has mentioned, if we just wanted to view stuff on screen, there would be no point to have increased resolution beyond about 1024*768, or whatever most people use on their monitors. You are not increasing the resolution of your monitor when you downsize your image - you are still viewing the same resolution.
Remember, there are two seperate "resolutions": the physical size (in pixels) of your image, and the resolution of the medium that's displaying it (eg, 1024x768 on many monitors).
When you "shrink" the picture down to fit on the whole screen, you are actually NOT making the pixels smaller. You're actually loosing detail - lots of it. But since the detail you are loosing is not there in the first place in the case of these sample images, the image looks better since it's being viewed at a size where the details look normal, instead of just an enlarged 2MP image.
What you are seeing on the monitor with a picture fit to the screen can convey no more detail than a 100% view can since the monitor is set at a fixed resolution of roughly 72dpi (changes from one screen to the next).
A camera should be able to produce pictures which look as good at full size then they are fit to the screen.
So, yes, your massively MP camera is producing a fine 2MP image, but in reality, a 2MP image gets you nowhere past viewing it on a monitor or using it on the web.
Printing is a critical part of photography, and print keeps the detail of the image. You can print at any DPI you want, limited only to your tools and/or budget. You can print at 600dpi and keep every single detail - including every single artifact. Sure, when you stand back and look at the print as a whole the problems may not be noticeable, but inspect it with a magnifier, and they're just as bad as they were in the samples we've seen.
Want to print posters with that camera? Sure, you'd only be able to print them at a low DPI, but those problems are going to look terrible.
Want to print 8x10 crops? The problems are still there.
Printing smaller? Well, then you're probably safe.
the point is, for an 14MP camera, it's acting like it's 2MP. That's not good. The 1Ds sure doesn't act like a 2Mp or even a 6MP camera. It acts like you'd expect - even at 100% views on screen, no artifacts... nothing but detail.
However, I suspect the samples you have seen of the 14N are worse than what the camera can output. It looks like they've had noise problems, and using noise reduction has killed any fine detail the image had except in the high contrast areas. Hopefully the delays are going to result in those noise problems being fixed.
I hope the 14N turns out to be a jewel. Canon could use some harsh competition - so could everyone. dSLR prices are just outragous right now, at least for a guy without any cash.
Good luck you! Now stop worrying about megapixels an go take some pictures!
Hello
A lot of people simply don't understand what I am trying to say.
I'll try to make it clear: Imagine you have a camera with 10.000 *
10.000 resolution. You take a picture of a part of a footballfield:
Imagine that the football in the picture takes up 2000*2000 pixels
(it's not square, but for simplicity).
Now you display this picture in 1:1 on a screen with e.g. 1000*1000
resolution. (Pretty close to what we have). Half (in each
direction) the football is now taking up ALL the screen, and of
course looks bad (steps because of your screen resolution).
If you now increase the Screen Resolution to 3000*3000, the
football takes up 1/9 of the screen area, and now looks very good.
If you reduce the Screen 500 * 500, the 1/4 of the football (in
each direction) now fills the screen, and you are so "close up" to
it that you can't even tell what it is!
Thus, the Screen Resolution versus the Resolution of the Ppicture
DOES matter when you Study a picture.
The only thing that is fair, is to have a Monitor that matches the
Picture resolution and look at the picture in 1:1 at such a
monitor. For the Kodak 14n this would a Monitor with 3500 * 4000
pixels resolution. Most of us does NOT have such a monitor.
Thus: Comparing pictures from different cameras on the same
resolution Monitor is NOT FAIR to the high res cameras
Geir Ove
Rylee Isitt
Senior Member
As an addendum, that 2000x2000 football shouldn't look bad at full size. It should look just as wonderful aas any other image displayed on your monitor, regardless of whether you're in at 100% or out at 10%.
If your image looks awesome when you've zoomed out to view it all at once, that means your monitor is quite capable of displaying a beautiful image. If you're camera isn't capable of producing an image that's beautiful at full size, something is wrong.
Full views at 100% on screen of the d30, d60, s2, d100, and 1Ds look beautiful. Why don't the ones from the 14N?
That football should look great, from 100% all the way down to fit-to-screen. With a high-res camera like the 1Ds or the 14n, you should be able to take a picture of a car from across a parking lot using a wide angle lens, crop it out at 100%, and get a very nice image on-screen.
Imagine monitors don't exist, just forget about them. Imagine your picture goes directly from the camera to a high-quality printer.
High resolution should let you print larger prints, or make smaller crops. The 14N samples defies this logic.
Hopefully the 14N itself doesn't!
If your image looks awesome when you've zoomed out to view it all at once, that means your monitor is quite capable of displaying a beautiful image. If you're camera isn't capable of producing an image that's beautiful at full size, something is wrong.
Full views at 100% on screen of the d30, d60, s2, d100, and 1Ds look beautiful. Why don't the ones from the 14N?
That football should look great, from 100% all the way down to fit-to-screen. With a high-res camera like the 1Ds or the 14n, you should be able to take a picture of a car from across a parking lot using a wide angle lens, crop it out at 100%, and get a very nice image on-screen.
Imagine monitors don't exist, just forget about them. Imagine your picture goes directly from the camera to a high-quality printer.
High resolution should let you print larger prints, or make smaller crops. The 14N samples defies this logic.
Hopefully the 14N itself doesn't!
Steven Snyder
Well-known member
see my post as an answer under "Computer screen resolution and beautiful mathematical formula!"
Here is why a 13.mp image may look worse than a 6mp image
Steve Snyder
Here is why a 13.mp image may look worse than a 6mp image
Steve Snyder
EarlRutherford
Forum Enthusiast
It is just a larger image in height by width.
David SL
Senior Member
You know, I was thinking the same thing about your responses!
Oh gosh,I wasn't being literal...just illustrating that there exists minimum dimensions that you can print a digital image to (and they are larger and larger for increasing image pixel density..) where you can exceed a given print devices maximum dpi. (and by so doing subsample noise,moire,ca and detail! in the image into oblivion.). For example, a 14n image printed at 8 x 10 would require 4500/10 = 400dpi along length and 3000/8 = 375dpi along width of real resolution from the printer to render all the provided image data at that print size , obviously if you only have 300dpi available from the printer, subsampling will effectively (it's not active interpolation..it simply occurs because the printer is unable to provide the required print resolution at the chosen print dimensions) occur during printing. Of course, most of the people who buy the 14n aren't looking to print 4 x 6 prints (all they'd need is a
Sure, the original Kodak samples right? the sucky ones? I don't care about those...if Kodak releases the camera with that kind of quality they deserve to go bankrupt..I am hoping that the Japanese shots (which are eons better than the Kodak shots) are more representative of what will be released. (crossing fingers)
Agreed, and as I stated before I never claimed the "scaling effect" had anything to do with the percieved quality of the original Kodak images, they indeed were lousy.
Peace,
--
David SL
Senior Member
In case you decide to pick at this nit.
whew!
--
screen image
My point was always that many artifacts viewable at 1:1 on screen (noise,ca,moire) would appear in print reduced(subsampled away) as you make smaller and smaller prints from the original data!
whew!
--
Rylee Isitt
Senior Member
Well, the only real reason is Bayer interpolation. 100% views are often softer than reduced views.
I'm not sure if you are the same guy who posted the manu formula site. I gave it a read over.
It makes sense in the context of why one camera produces a larger image than another (AKA common sense), but has nothing to do with quality loss. A pixel is a pixel.
If your camer's sensor has 2 million pixels, you can view it at full size and it looks great. Same goes for a camera that has 2 billion (you might be resolving microscopic detail at that point, though, but it will still be detail).
I'm not sure if you are the same guy who posted the manu formula site. I gave it a read over.
It makes sense in the context of why one camera produces a larger image than another (AKA common sense), but has nothing to do with quality loss. A pixel is a pixel.
If your camer's sensor has 2 million pixels, you can view it at full size and it looks great. Same goes for a camera that has 2 billion (you might be resolving microscopic detail at that point, though, but it will still be detail).
g.wilmshurst
New member
- Messages
- 9
- Reaction score
- 0
Oh great ! It is really nice to know that I could screw my Zeiss 85mm 1.4
to a 2 billion pixel camera and it would resolve detail down to one pixel!!!!!!!!!!!!!!!!!!!!!!!!!! !!
i WILL PASS THIS ON TO ZEISS THEY MIGHT AS WELL SHUT RESERCH AND DEVELOPMENT DOWN AS THE PERFECT LENS HAS ALREADY BEEN DEVELOPED.
Regards
graham
Rylee Isitt
Senior Member
I do hope you realise it was a hypothetical situation. Of course a lens can't resolve microscopic detail. Maybe I would have been better off using something closer to 14MP, but I felt an extreme was needed...
When a 2 billion MP sensor is avaiable, I'm sure that will be a few hundred years from now when lenses are just as "perfect"
When a 2 billion MP sensor is avaiable, I'm sure that will be a few hundred years from now when lenses are just as "perfect"
g.wilmshurst
New member
- Messages
- 9
- Reaction score
- 0
Of course I doI do hope you realise it was a hypothetical situation. Of course a
lens can't resolve microscopic detail. Maybe I would have been
better off using something closer to 14MP, but I felt an extreme
was needed...
When a 2 billion MP sensor is avaiable, I'm sure that will be a few
hundred years from now when lenses are just as "perfect"
The point I am making is that as pixel counts increase other factors are coming into play ie: lens quality mirror slap focus accuracy and electronic noise etc.
Tis will mean that say a 20 mpixel camera will not have the Theoretical increase in resolution over a 10 mpixel camera.
Graham
Rylee Isitt
Senior Member
I wholeheartedly agree.
The camera blur on such high detail would be significant even at high shutter speeds. This is, of course, assuming no major changes in lens/shutter technology.
I guess you could bolt a 2 ton tripod to the ground and mount a 100 pound camera on it to prevent camera shake...
All this assuming you ever have a 2 billion pixel sensor, that is
The camera blur on such high detail would be significant even at high shutter speeds. This is, of course, assuming no major changes in lens/shutter technology.
I guess you could bolt a 2 ton tripod to the ground and mount a 100 pound camera on it to prevent camera shake...
All this assuming you ever have a 2 billion pixel sensor, that is
Of course I doI do hope you realise it was a hypothetical situation. Of course a
lens can't resolve microscopic detail. Maybe I would have been
better off using something closer to 14MP, but I felt an extreme
was needed...
When a 2 billion MP sensor is avaiable, I'm sure that will be a few
hundred years from now when lenses are just as "perfect"
The point I am making is that as pixel counts increase other
factors are coming into play ie: lens quality mirror slap focus
accuracy and electronic noise etc.
Tis will mean that say a 20 mpixel camera will not have the
Theoretical increase in resolution over a 10 mpixel camera.
Graham
g.wilmshurst
New member
- Messages
- 9
- Reaction score
- 0
Anyway back to reality I really hope that when the kodak comes out that its pics are as much improved over the japanese pics as the Japanese pics are over kodaks first release (sorry its late over here)I wholeheartedly agree.
The camera blur on such high detail would be significant even at
high shutter speeds. This is, of course, assuming no major changes
in lens/shutter technology.
I guess you could bolt a 2 ton tripod to the ground and mount a 100
pound camera on it to prevent camera shake...
All this assuming you ever have a 2 billion pixel sensor, that is
For approx twice the cost of an F5 it would be very tempting,No processing costs and not having to shove slide after slide into my nikon 4000 ed would make it a steal
Well, I think that is theoretically impossible for visible wavelengths given a sensor the size of a 35mm frame. You're talking about individual sensor elements smaller than the wavelength of visible light. In such a situation, the light would simply pass through the sensor.When a 2 billion MP sensor is avaiable, I'm sure that will be a few
hundred years from now when lenses are just as "perfect"
Jack40001
Leading Member
Appear? Huh? Or disappear?
you make smaller and smaller prints from the original data!
whew!
--
R
Rinus Borgsteede
Guest
I agree that there is no magnification but the image is quite large and looking at 100% is a fair method of inspecting the quality.
I suggest that looking at an 8x10 of a 35mm camera and a 16x20 of a 6x7 negative is roughly the same enlarging ratio but we never look at these ways of comparing because it would be expensive and irrelevant as images would not be viewed at the same distance. The only way to compare would be to restrict the size of the print and hold similar prints side by side for comparisson.
If we held similar prints side by side, the full monitor screen would be a good way to start but because screen resolution is quite low, many cameras would look very much the same as long as the same pixel size was used for the on screen comparisson. Sqeezing hi res. images into a downsampled screen would indeed make all cameras much the same.
Viewing the various images pixel by pixel on a screen is an unfair test but the method I described is even more unfair.
Ultimately, the final tests is not the image on the screen but the the findings of real practical use in a real world of digital output into print.
Most of us want the highest resolution but most of us will never use that camera that way because of the large file size. In my opinion, the advantage with the N14 is the full size CMOS chip not the enormous file it produces. I just went to a seminar and the portrait shooters are happy as it is with a 2000x3000 pixel image. they do enough retouching as it is and it takes time to convert all these images.
Rinus of Calgary
Keyboard shortcuts
- f
- Forum