corndog wrote:
Uprezzing a 12mp image to 24mp in order to compare it to a 24mp image doesn't sound fair because half the data originates from a processor instead of photons.
WillWeb wrote:
I shoot birds with a 1DmkII and 500/4... ...When I start shooting with a 7D, I expect to be able (when needed) to crop a lot tighter than I can now, because of the higher pixel density...
That is not a comparison. You are intending to use the 7D in a completely different way to the way you use your 1D2.
My suggestion is that uprezzing and viewing at 100% and equal magnification is a way of COMPARING noise and image quality between cameras which have different numbers of pixels. It is almost impossible to view two files at 100% and not compare them mentally. Where the two files have different numbers of pixels, that comparison is in error. That is why one must be extremely careful to equalise magnification before comparing, and that is also why many people groan unnecessarily when they read that the 7D has 18 Mpixels. They are misled by an illusion every day, using their own cameras and looking at review sites.
this whole issue of downrezzing a 21mp to a 12mp to dertermine noise,DR ect
seems like JUNK science kinda like the global warming thing..when you know hypocrites on the Left saying the right wants to pollute and destroy the planet,
but in turn the left backs the coal industry by labeling it Clean coal.
digitalbug30d wrote:
the top one is the uprezzed one
Correct. Note that it has been uprezzed AND downrezzed. I don't know if those two processes are equally culpable in the almost invisible loss of per pixel contrast. Bear in mind that the loss of detail is at least an order of magnitude smaller than the effect of read noise on detail at high isos. What that means is that the 12 Mpixel file will essentially not be handicapped in a comparison at 21 or 24 Mpixels when the iso is higher than 200-400. The loss of detail in the iguana is on a much smaller scale than the large and lumpy noise artefacts that affect files at 800 iso and upwards.
brainiac wrote:
That is not a comparison. You are intending to use the 7D in a completely different way to the way you use your 1D2.
Well, not completely different, but yes the reason I want a high-density sensor is to do things I can't with my current equipment. Otherwise 8 MP would be enough.
I think we are in agreement, just using different words.
corndog wrote:
depends on the amount of uprezzing and ones ability to differentiate...
brainiac wrote:
Uprezzing isn't very destructive to your image.
I stand by my comment, and your comment about uprezzing is generalized. If you want to add criteria to include the changes you've made, I would agree that it isn't very destructive. One can't agree or disagree to your comment as is because there are no parameters.
Let's make this even simpler, let's compare this 20p photo to a 40p photo. The 20p photo had to duplicate its data to become 40p, but the real life data was different, and the 40p sensor was able to capture it, which was captured and which was duplicated?
corndog wrote:
I stand by my comment, and your comment about uprezzing is generalized. If you want to add criteria to include the changes you've made, I would agree that it isn't very destructive. One can't agree or disagree to your comment as is because there are no parameters.
The level of destruction to the file by this _double_ interpolation is practically negligible, and certainly well below the threshold needed to compare high iso noise in two files. That's why the uprezzing method of comparing crops from two cameras with different numbers of pixels is a fair method. Uprezzing does not significantly change the content of the lower Mpixel file.
>Let's make this even simpler, let's compare this 20p photo to a 40p photo. The 20p photo had to duplicate its data to become 40p, but the real life data was different, and the 40p sensor was able to capture it, which was captured and which was duplicated?
Uprezzing is creating "data" that does not exist in the RAW file. and is not part of the image.
view the full size images onscreen the way your would when normally viewing from both cameras - then decide which one looks better onscreen - both reduced to fit the screen and at actual size
make prints from both at the size you would normally - then decide which one gives the better print,
A real world test of how prints look and how full size images look viewed on screen.
You used your example, which had specific criteria, as a supporting statement to uprezzing not being very destructive. Here, look..I uprezzed from 3000x2000 to 3002x2002 and it looks great, therefore uprezzing is not very destructive! I used my example to show you the extreme opposite of that. I'm saying that uprezzing does not give you real data, and could give errant data compared to what a pixel might pick up.
I think we're going to have to agree to disagree. Just please keep in mind that people are not wrong because they don't use the same criteria and reference points as you do for evaluating image quality.
corndog wrote:
You used your example, which had specific criteria, as a supporting statement to uprezzing not being very destructive. Here, look..I uprezzed from 3000x2000 to 3002x2002 and it looks great, therefore uprezzing is not very destructive! I used my example to show you the extreme opposite of that. I'm saying that uprezzing does not give you real data, and could give errant data compared to what a pixel might pick up.
I think we're going to have to agree to disagree. Just please keep in mind that people are not wrong because they don't use the same criteria and reference points as you do for evaluating image quality....Show more →
Please show me an actual uprez that illustrates the phenomenon in your diagram. I've never seen bicubic interpolation behave like that.
The computer does not invent new data when interpolating, it works out, mathematically and reliably, what the old dataset would look like in a different number of pixels. If that kind of calculus satisfied Leibniz, Newton, and Einstein, then I don't think I'm going to pay too much heed to people's skepticism towards infinitessimals. The interpolation process works and is extremely accurate, as my example demonstrates. The idea that mystery data corrupts your image when you interpolate it is just tin-foil-hat mumbo-jumbo. Here in the practical world uprezzing gives you a level playing field to compare cameras without always having to make same-sized prints.
The dataset has already been interpolated once when it was demosaiced, and if you want to compare prints it will be interpolated again into ink drops. Interpolation is nothing to be scared of. You might just as well avoid using motors because you don't believe in magnets.
It invents it via mathematical equation! If it's so accurate, why is it we can see the difference with your example? Your own example disproves your own point that these differences are infinitesimal.
Listen, I understand your points, and with parameters you are correct. Do you not understand my points, or just disagree to the level of impact? My disagreement with your points has been said by me over and over, you need criteria to support your statements. You can't accurately say uprezzing isn't very destructive unless you give some parameters. I can downsize a photo to 1 pixel, and uprez it to 1000 if you really want me to!
brainiac wrote:
Let's make this simpler. One of the following images has been uprezzed to 1750 pixels wide, and then downrezzed back to 1000 pixels wide and sharpened slightly. Which is the uprezzed and downrezzed one, and which the original?
Good luck...
Uprezzing isn't very destructive to your image.
You had to sharpen it slightly and how slight is that "slightly"?
Sizing up 1000px to 1750 is not the same as sizing up 4368 x 2912 to 5616 x 3744.
Besides, I don't think uprezzing 12 to 24 makes for fair comparison. As abqnmusa said, you're adding pixels that weren't there in the first place. Downrezzing 24 to 12 is closer to being fair, it throws pixels away but then 24mp would theoretically have double the details anyway.
If uprezzing had no impact on image quality we could save money and use a 3 MP or 6 MP camera. Just uprez it up to 21 MP. Of course that would not work.
The only way to compre is to view full size images from each camera onscreen and in prints. If there is a difference you will see it.
Uprezzing with a nearest neighbor filter is guaranteed to do nothing to your original image (all pixels in the original image will be exactly preserved in the uprezzed image). The resulting uprezzed image will look like @$$ though. Perhaps we should compare with compare uprezzed images with a NNB filter instead
corndog wrote:
It invents it via mathematical equation! If it's so accurate, why is it we can see the difference with your example? Your own example disproves your own point that these differences are infinitesimal.
Listen, I understand your points, and with parameters you are correct. Do you not understand my points, or just disagree to the level of impact? My disagreement with your points has been said by me over and over, you need criteria to support your statements. You can't accurately say uprezzing isn't very destructive unless you give some parameters. I can downsize a photo to 1 pixel, and uprez it to 1000 if you really want me to! ...Show more →
I have never claimed that downrezzing is not destructive. That is the source of your confusion. Uprezzing and downrezzing are very different. Downrezzing discards information, whereas uprezzing doesn't. I have shown that an uprez followed by an equivalent downrez produces very nearly no effect except a difference in sharpness. That's what the maths predicts and that's the result you get.
