Up-Rez and down-rez for comparisons - why?

[Deardorff]

Am wondering why comparisons of digital cameras have to be up or down rezzed to make a comparison?

With film if I wanted to make a comparison I made prints of the same size from the same of similar film. Comparing a 35mm negative to a 120 negative to 4x5 negative to an 8x10 contact print was easy to so. Printed each to 8x10 and set side by side and looked. Got closer or magnified if needed.

Same with 16x20 size prints from various sized negatives.

Could then do them developed in specific developers or even printed on specific papers if we wanted.

Why isn't it that easy with digital?

[ErikKaffehr]

Hi,

If you print any image it will be resized either in your editor or in the printing process.

Let's say we talk about 16x20" prints from my two cameras printed on Epson SP 3880 at 360 PPI:

16" at 360 PPI is 5760 pixels

The Sony Alpha 99 has 4000 pixels vertically, so it will be uprezzed 44% for a 16x20" 36PPI print

The P45+ has 5433 pixels vertically, so it will be upprezzed 6%

Now, assume you have 1" borders on that 16x20" so it will get 14"x18". In that case 5040 pixels are needed vertically at 360 PPI.

So, in this case the Sony Alpha 99 image would be uprezzed 26% and the P45+ would be downrezzed 8%.

Clearly, the printer driver will do it for you, but resizing algorithms in Photoshop or Lightroom are better than the ones in the printer driver.

That said, I would say that my experience is that there is very little difference between 24MP and 39MP in A2 size (16 x 23") prints.

Best regards
Erik

Am wondering why comparisons of digital cameras have to be up or down rezzed to make a comparison?

With film if I wanted to make a comparison I made prints of the same size from the same of similar film. Comparing a 35mm negative to a 120 negative to 4x5 negative to an 8x10 contact print was easy to so. Printed each to 8x10 and set side by side and looked. Got closer or magnified if needed.

Same with 16x20 size prints from various sized negatives.

Could then do them developed in specific developers or even printed on specific papers if we wanted.

Why isn't it that easy with digital?

[MarkL]

Contact prints from film formats will all be different sizes so to compare them you need an enlarger and a lens to enlarge the image like you suggest. Since printing from digital is not an optical process if different cameras produce files with different amounts of pixels they need to be resampled to all produce the same prints size (or even size on a screen at 100%).

[dwswager]

Am wondering why comparisons of digital cameras have to be up or down rezzed to make a comparison?

With film if I wanted to make a comparison I made prints of the same size from the same of similar film. Comparing a 35mm negative to a 120 negative to 4x5 negative to an 8x10 contact print was easy to so. Printed each to 8x10 and set side by side and looked. Got closer or magnified if needed.

Same with 16x20 size prints from various sized negatives.

Could then do them developed in specific developers or even printed on specific papers if we wanted.

Why isn't it that easy with digital?

The big problem with that is how do you show a print to your online audience? 

In a way, since lower resolution is more standard than higher, by downreszing to a common size, you can compare two cameras at equal output.  If larger size is the target, then I think upsampling to a common size would be more indicative.  But you are correct in general that larger film and larger and higher pixel count sensors both have an advantage and that get subverted in a downsampling scheme.

[Telecaster]

You can indeed compare the results of different electronic cameras/sensor formats via print. My approach is simply to make the best 15x20" or 14x21" print I can from each camera/lens config, then place the prints side by side and see how they look. I don't try to equalize pixel counts prior to printing, but I do make sure I'm putting at least 300ppi on the paper. This means sometimes up-rezing a bit.

The most significant things I've learned from doing this are: 1) All my current gear can handle 15x20/14x21" with ease; and 2) All my current cameras, when fitted with top quality lenses, can resolve spatial detail that can't be seen in a 15x20/14x21" print from my Epson 3880. This happens to be my largest print size aside from the rare ~15x30" pano. 

-Dave-

[ErikKaffehr]

Hi,

According to our printing guru, Jeff Schewe it is best to resample to 360 PPI before printing on Epson as it is the native resolution of the printer. If the resolution is different, it will be resized by the printer driver, which is using a less good algorithm. Canon printers have 300 PPI native resolution.

If the resolution of the camera is above 360 or 300 PPI, than it can be useful to use 720 PPI (Epson) or 600 PPI (Canon) and use "Superphoto" resolution.

This may be splitting hairs of course. I don't think I can see the difference without a loupe.

Best regards
Erik

You can indeed compare the results of different electronic cameras/sensor formats via print. My approach is simply to make the best 15x20" or 14x21" print I can from each camera/lens config, then place the prints side by side and see how they look. I don't try to equalize pixel counts prior to printing, but I do make sure I'm putting at least 300ppi on the paper. This means sometimes up-rezing a bit.

The most significant things I've learned from doing this are: 1) All my current gear can handle 15x20/14x21" with ease; and 2) All my current cameras, when fitted with top quality lenses, can resolve spatial detail that can't be seen in a 15x20/14x21" print from my Epson 3880. This happens to be my largest print size aside from the rare ~15x30" pano. 

-Dave-

[Bart_van_der_Wolf]

Hi,

According to our printing guru, Jeff Schewe it is best to resample to 360 PPI before printing on Epson as it is the native resolution of the printer. If the resolution is different, it will be resized by the printer driver, which is using a less good algorithm. Canon printers have 300 PPI native resolution.

If the resolution of the camera is above 360 or 300 PPI, than it can be useful to use 720 PPI (Epson) or 600 PPI (Canon) and use "Superphoto" resolution.

This may be splitting hairs of course. I don't think I can see the difference without a loupe.

Hi Erik,

While the differences may be subtle at times (depends on the subject matter), an often overlooked fact is that 'Output' sharpening at the 'native' output resolution of the printer, i.e. after all resampling, will give better results. Also, 720 or 600 PPI allows to do much more accurate sharpening of the (usually low contrast) detail that is there, and can be stronger on larger detail without creating artifacts that may become visible. There are simply more pixels available to dither the details with, and they will not be resampled (which often loses contrast) anymore.

The effect of proper resizing on printed output can range from subtle, to dramatically better, especially when one uses a tool like Topaz Detail (which can additionally also augment the visual effect at larger viewing differences).

Cheers,
Bart

[dwswager]

Hi,

According to our printing guru, Jeff Schewe it is best to resample to 360 PPI before printing on Epson as it is the native resolution of the printer. If the resolution is different, it will be resized by the printer driver, which is using a less good algorithm. Canon printers have 300 PPI native resolution.

I certainty don't want to try to speak for him, but based on Jeff's comments on LuLu and articles this is incorrect.  Jeff seems to accept Epson Representative's statements that the printer driver does not resample data.  He believes, but has not been able to confirm, that a component of the Mac and Windows operating system resamples the data.  At the end of the day, it is immaterial as most people agree that 360ppi produces the best looking output (in most cases) from Epson Stylus printers.  Knowing the limited functionality of Windows GDI, I don't think the data is resampled at all and personally think the 360ppi comes from Epson's original statement that the dithering algorithm operates on a macroblock of 60x60 pixels and tops out, except for specific types of images, at 360ppi.

The other issue is that Epson printers do not have a native resolution in the classic sense.  They are not continuous tone, but dithering printers.  They actually have the ability to put down a lot more dots than 360 per inch, but due to the droplet size and spread, they can't use every dot location available to them. This is where the 1440x720 and 2880x1440 numbers come from. But unlike a continuous tone printer they can use the extremely high dot locations available and varying dot sizes to create a pattern to simulate continuous tone from 4, 6, 7, 8, 9, 10 different colors.

http://www.digitalphotopro.com/technique/photography-workflow/the-right-resolution

Printer Output Resolution

Today's high-end, large-format inkjet printers are capable of outputting at high resolution. The Epson pro printers' print heads have a reported output resolution of 360/720 dpi, depending on the print mode. Canon and other printers that use similar thermal head technology have a reported 300/600 dpi. What do I mean by "reported"? The print driver communicates with the operating system print pipeline and states its resolution in dots per inch. You can send any output resolution to the print head, but the print pipeline will resample the input resolution and send the printer the resolution it asks for. Now there's some question of where and how this data gets resampled.

The people at Epson say that the print driver doesn't do the re-sampling, and since the application sending the print doesn't do the resampling unless asked to do so, the resampling must be happening in the print pipeline. I've tried, unsuccessfully, to get confirmation from Apple and Microsoft about their respective print engine activities regarding the resampling of the image data. So I really don't know where or how the resampling is being done. But I'm convinced some sort of resampling is being done. Is it an optimal resampling algorithm, or is it something done for speed? I don't know, but I suspect, at best, it's a compromise in favor of speed. I'm pretty sure there are better, optimized resampling algorithms that could do a superior job. In fact, Adobe Photoshop Lightroom resampling is a hybrid Bicubic algorithm that interpolates between Bicubic and Bicubic Smoother for upsampling and Bicubic and Bicubic Sharper for downsampling.

[Robcat]

I've been wondering same thing as OP for some time---and conclude that he is correct. The business most responders have been going into about what the printer drivers do and don't do are beside the point, at least beside the point of Deardorff's comment, which concerned down-rezzing high MP images to compare them to lower MP images. I'm not interested in pixel performance  nearly as much as I am print performance.
Say I like to shoot full length portraits and print them at 20 x 40 (which I do). I want to see how sensor A compares to sensor B when I look at 2 shots framed the same way (head to toe, model standing covering most of the sensor) and printed at the same size. THAT's what tells me one gives me better output than the other (of course assuming I controlled for lens quality). It makes no sense to show me a shot of the model's nose and compare that to a shot of her entire face because "they both have the same number of pixels." Or to show 2 shots of her face with one down-rezzed to give the same number of pixels.
Do what Deardorff suggests (and I've long done) and take the same shot, print it at the same size and look at the two.
The issue of "how do I show the results online" is a separate issue. But it's the same problem as how to show other subtleties in an uploaded .jpg.

[dwswager]

I've been wondering same thing as OP for some time---and conclude that he is correct. The business most responders have been going into about what the printer drivers do and don't do are beside the point, at least beside the point of Deardorff's comment, which concerned down-rezzing high MP images to compare them to lower MP images. I'm not interested in pixel performance  nearly as much as I am print performance.
Say I like to shoot full length portraits and print them at 20 x 40 (which I do). I want to see how sensor A compares to sensor B when I look at 2 shots framed the same way (head to toe, model standing covering most of the sensor) and printed at the same size. THAT's what tells me one gives me better output than the other (of course assuming I controlled for lens quality). It makes no sense to show me a shot of the model's nose and compare that to a shot of her entire face because "they both have the same number of pixels." Or to show 2 shots of her face with one down-rezzed to give the same number of pixels.
Do what Deardorff suggests (and I've long done) and take the same shot, print it at the same size and look at the two.
The issue of "how do I show the results online" is a separate issue. But it's the same problem as how to show other subtleties in an uploaded .jpg.

I absolutely agree that comparing the sensor data in the final destination output medium (specific paper, printer, ink set) and size is the only really valid method of comparison.  However, most online publications have varried audiences and since smaller print size is by far the most common, then it makes some sense for them to down-sample as long as down-sampling is limited to the extent that still encompasses the most common resolution needs.

As to presentation, I know what I can do with say 22, 24, 36 and 50 MP 2:3 captures.  Just show me actual pixels of key areas of both imgages such as the shadows, specular highlights or significant patterns or colors. We already have the specific lens used and raw conversion (in camera or external) in the mix.

[robdickinson]

Am wondering why comparisons of digital cameras have to be up or down rezzed to make a comparison?

Basically because viewing on screen at 100% an 8mp file vs a 36mp file is stupid, and too many people got into the hang of that when resolution was very similar, then got 15+mp cameras and OMG 100% view looks WORSE HOW CAN THEY DO THIS TO ME!!! moments.

Printing equal size is a good test esp if it is your normal print size/workflow, but this isnt very web friendly..

As to what I do, it depends on what the intent is.

I downsampled my 6d astro frames to 12mp to compare with a sony a7s file.

Whilst for base ISO landscapes I might upsample it to compare against the a7r

[Schewe]

I certainty don't want to try to speak for him, but based on Jeff's comments on LuLu and articles this is incorrect.

Well, I'll speak for me...the print pipeline on both Mac and Windows is not easily dissected. But one can prove to oneself by doing actual tests. Epson prints at either 360PPI or 720PPI depending on the print settings. If you upsample to 360PPI (if the native rez is less than 360PPI) one can see the result...less aliasing effects with circles and diagonal edges. Same deal if the native rez is above 360PPI but below 720PPI (and you have an Epson pro printer with Finest Detail in the driver). Your results will be better if you upsample in LR or Photoshop to 720PPI and then sharpen.

That's really the key here...do you allow the print pipeline to resample or do you do a proactive upsample then sharpen at the final print size? If you let the print pipeline do the resampling (and use at best Bilinear and prolly Nearest Neighbor) your results are less good. That's easy to prove just by doing prints at various resolutions.

And yes, I STILL don't know where in the print pipeline the resampling is being done but it's pretty obvious that resampling is being done somewhere. The guys at Epson that tell me it ain't the print driver is believable (regardless of what others have said). But resampling is being done somewhere if the reported rez is not either 360/720 PPI (depending on the driver settings).

Personally, I'm done with the testing...I've proven to myself what I say is correct. You can say whatever you want to say but in my mind, unless you prove it, you are just flapping you're gums. If you image is less than 360PPI at "native resolution" resample to 360PPI and then output sharpen. If your native rez is between 360 to 720, select Finest Detail and resample to 720PPI and then output sharpen. If you don't believe me, test it yourself...if your results are other than mine, I'll eat my hat! (that's how sure I am at saying what I say–I don't eat hats).

[Bart_van_der_Wolf]

Well, I'll speak for me...the print pipeline on both Mac and Windows is not easily dissected. But one can prove to oneself by doing actual tests. Epson prints at either 360PPI or 720PPI depending on the print settings.

I agree with Jeff on this one. The closest we all can come to understanding what's happening, is by reverse engineering what's going on based on what we can see in the printed result.

If you upsample to 360PPI (if the native rez is less than 360PPI) one can see the result...less aliasing effects with circles and diagonal edges. Same deal if the native rez is above 360PPI but below 720PPI (and you have an Epson pro printer with Finest Detail in the driver). Your results will be better if you upsample in LR or Photoshop to 720PPI and then sharpen.

That's really the key here...do you allow the print pipeline to resample or do you do a proactive upsample then sharpen at the final print size? If you let the print pipeline do the resampling (and use at best Bilinear and prolly Nearest Neighbor) your results are less good. That's easy to prove just by doing prints at various resolutions.

It's as simple as that, and the same goes for Canon and HP printers, at 300 and 600 PPI. It's all a result of the number of ink nozzles available in the print-head, the positioning accuracy of the print-head and the paper feed mechanism. Add to that that it is much easier to design a dither pattern for blending color nuances between the pure ink colors at two specific resolutions, and the observed effects on print quality, and it's hard to draw any other conclusions based on the empirical evidence.

And yes, I STILL don't know where in the print pipeline the resampling is being done but it's pretty obvious that resampling is being done somewhere. The guys at Epson that tell me it ain't the print driver is believable (regardless of what others have said). But resampling is being done somewhere if the reported rez is not either 360/720 PPI (depending on the driver settings).

There seems to be a difference between the Mac OS and Windows, where the Mac based driver delegates the part of the resampling to some built in OS functionality before feeding the final/resampled data to the printer after dithering. In the Windows driver versions it seems to be a function of the printer driver itself, but nobody really knows besides the Epson/Canon engineers.

Personally, I'm done with the testing...I've proven to myself what I say is correct. You can say whatever you want to say but in my mind, unless you prove it, you are just flapping you're gums. If you image is less than 360PPI at "native resolution" resample to 360PPI and then output sharpen. If your native rez is between 360 to 720, select Finest Detail and resample to 720PPI and then output sharpen. If you don't believe me, test it yourself...

I agree, and for that purpose I've created a printer test target (I'm in the testing phase). The target may be too cruel for average use, but it should show any loss of resolution or resampling artifacts that the print process may generate. Most likely, besides profile conversion issues, the print head alignment and imprecise paper feeding will also add some artifacts, as does ink diffusion in the print medium. I'll be happy to discuss them in a separate thread.

Here are the current targets for the masochists amongst us:
Test target for 600 PPI printers
Test target for 720 PPI printers

The numbers on the targets are calibrated for resolution expressed in cycles/mm when the prints are done at 600/720 PPI respectively (which produces a square target size of 130mm). For the moment it should be enough to know that 5 cycles/mm delivers a very good print quality at reading distance under normal lighting conditions, and 8 cycles/mm would produce an excellent print quality. The printers are capable of even better results, but that can only be exploited for real images with very good output sharpening.

So to get back to this thread's topic, and answer the OP's question, if you don't resample different sized files then they either display/print at different output sizes, or they will get resampled behind the scenes, and both make it harder/impossible to really compare image quality.

Cheers,
Bart

[Hulyss]

Am wondering why comparisons of digital cameras have to be up or down rezzed to make a comparison?

I don't know either. I see people doing that only on internet thought (comparisons).

With film if I wanted to make a comparison I made prints of the same size from the same of similar film. Comparing a 35mm negative to a 120 negative to 4x5 negative to an 8x10 contact print was easy to so. Printed each to 8x10 and set side by side and looked. Got closer or magnified if needed.
Same with 16x20 size prints from various sized negatives.
Could then do them developed in specific developers or even printed on specific papers if we wanted.
Why isn't it that easy with digital?

If you do it with digital, differences start to show up only around A2. You tend to soften little prints (because people see them closely) and pull out details via different methods for bigger prints. It is not that easy with digital for someone who have no notions into printing. Exactly like with film if someone have no notion into enlarging. So now, photographers can have very pro printing solutions at home and then need to be photographer AND printer which are two very different jobs. A professional printer also have an eye. His eye is more about color perception and proportions than the photographer eye. With film it is more easy because sharpening is based only on contrast level and film grain (or size).

[Bart_van_der_Wolf]

I don't know either. I see people doing that only on internet thought (comparisons).

It's mainly to avoid the errors/artifacts that can be created behind the scenes by (low quality) resampling algorithms. We'd want a level playing field that only differs in captured image quality, not caused by other factors.

So it's all about reducing the number of unknown variables. Unfortunately, the resampling itself can also cause issues, so it's not a simple thing to compare image quality unless one has a good grip on the total workflow.

Cheers,
Bart

[dwswager]

Personally, I'm done with the testing...I've proven to myself what I say is correct. You can say whatever you want to say but in my mind, unless you prove it, you are just flapping you're gums. If you image is less than 360PPI at "native resolution" resample to 360PPI and then output sharpen. If your native rez is between 360 to 720, select Finest Detail and resample to 720PPI and then output sharpen. If you don't believe me, test it yourself...if your results are other than mine, I'll eat my hat! (that's how sure I am at saying what I say–I don't eat hats).

This is the real point.  While we would all like to know definitively what is happening with our data, the real issue is what gives  you the best output.  Use advice, especially from experienced people like Jeff, to print a few samples and decide for yourself what works best in your situation!

[Jack Hogan]

Basically because viewing on screen at 100% an 8mp file vs a 36mp file is stupid, and too many people got into the hang of that when resolution was very similar, then got 15+mp cameras and OMG 100% view looks WORSE HOW CAN THEY DO THIS TO ME!!! moments.

Right.  In such a situation it may be useful to suggest downsizing the larger image to the resolution of the smaller one (or vice versa upsizing the smaller one) with a common, decent algorithm and re-viewing the resulting images at 100% or however.  Is this what the OP was asking?

[ErikKaffehr]

Hi,

Printing large may be the most adequate comparison. But, pixel peeping will show differences less obvious in print. Or, rather a loupe is needed to observe the differences in print.

Making comparisons on screen is more economical than making large prints. One approach I would suggest that both images are upsized to a given print size at say 180 PPI. Most screens are around 100 PPI so looking at the images at actual pixels gives around 1.8x enlargement, but computer screens are not intended for really short viewing distances. By and large,I guess that 180 PPI studied at actual pixels on screen will be a good estimate of pixel peeping a large print. A good algorithm should be used for resizing, the one in Lightroom is pretty good, much better than Photoshop "bicubic", IMHO.

Personally, I have seen amazingly little difference between 12 and 24 MP in A2 size prints and also amazingly little difference between 24 MP (full frame) and 39 MP (MFD) in A2 size. In the 24MP/39MP comparison I am pretty sure the difference is readily observable at A1 size (23" x 34").

Best regards
Erik

Right.  In such a situation it may be useful to suggest downsizing the larger image to the resolution of the smaller one (or vice versa upsizing the smaller one) with a common, decent algorithm and re-viewing the resulting images at 100% or however.  Is this what the OP was asking?

[dwswager]

Hi,

Printing large may be the most adequate comparison. But, pixel peeping will show differences less obvious in print. Or, rather a loupe is needed to observe the differences in print.

Making comparisons on screen is more economical than making large prints. One approach I would suggest that both images are upsized to a given print size at say 180 PPI. Most screens are around 100 PPI so looking at the images at actual pixels gives around 1.8x enlargement, but computer screens are not intended for really short viewing distances. By and large,I guess that 180 PPI studied at actual pixels on screen will be a good estimate of pixel peeping a large print. A good algorithm should be used for resizing, the one in Lightroom is pretty good, much better than Photoshop "bicubic", IMHO.

Personally, I have seen amazingly little difference between 12 and 24 MP in A2 size prints and also amazingly little difference between 24 MP (full frame) and 39 MP (MFD) in A2 size. In the 24MP/39MP comparison I am pretty sure the difference is readily observable at A1 size (23" x 34").

Best regards
Erik

The bottom line answer as to why it is done is that it is easy and cheap and gives a standard for comparison.  But, the question becomes "Is that standard of comparison valid for my your intended purpose(s)?"   I think in a lot of cases it is.  But printing big is one of those cases in which a different standard is required.  Every comparison tells you something, but only you can decide if what it is telling you is meaningful.

Even for my sports shooting that is mostly destined for screen, it might not be as simple as it sounds.  When shooting at night, I need fast lenses, but don't own anything extremely long in the focal length while extremely fast in aperture.  So I need to compare a crop from a 36MP Full Frame vs 24MP Full Frame vs 24MP APS-C.  I'm balancing the magnification versus the image quality (mainly less noise from the FF cameras).

[Jack Hogan]

Even for my sports shooting that is mostly destined for screen, it might not be as simple as it sounds.  When shooting at night, I need fast lenses, but don't own anything extremely long in the focal length while extremely fast in aperture.  So I need to compare a crop from a 36MP Full Frame vs 24MP Full Frame vs 24MP APS-C.  I'm balancing the magnification versus the image quality (mainly less noise from the FF cameras).

If you do not fill your sensor (e.g. BIF) usually the smaller pixel wins, all else equivalent.

Jack