Luminous Landscape Forum

Equipment & Techniques => Beginner's Questions => Topic started by: JohnTodd on September 16, 2010, 04:11:10 pm

Title: Why upres images for printing?
Post by: JohnTodd on September 16, 2010, 04:11:10 pm
Not wanting to hi-jack the recent thread about *how* to up-res befor printing, I would like to know *why*.

My thinking is this. I shot an image with a certain actual resolution, cropped it and edited the remaining 'camera' pixels. If I send those to the printer, the printer driver is going to map those pixels to actual nozzle-squirts as effectively as it knows how.

What is to be gained by increasing the resolution of the image in Photoshop (or similar) rather than letting the printer driver have maximum control over the mapping? Can Photoshop create a better interpolated pixel than the printer driver?
Title: Re: Why upres images for printing?
Post by: Bart_van_der_Wolf on September 16, 2010, 05:35:48 pm
What is to be gained by increasing the resolution of the image in Photoshop (or similar) rather than letting the printer driver have maximum control over the mapping? Can Photoshop create a better interpolated pixel than the printer driver?

Hi John,

Depending on the printer driver, yes, Photoshop can probably do better. There are other resampling methods that do even better. What's more, after the resizing to the required pixel density for the intended output size you can also compensate for the additional losses that will occur when the printer lays down its colors on the output medium. The result is sharper prints with fewer artifacts.

Cheers,
Bart
Title: Re: Why upres images for printing?
Post by: feppe on September 16, 2010, 05:42:46 pm
Mr Schewe advocates keeping the photo at native pixel dimensions for printing, as long as this results in a print between 180 and 360 dpi. Outside that PS, Qimage or dedicated resampling programs are recommended.
Title: Re: Why upres images for printing?
Post by: Bart_van_der_Wolf on September 16, 2010, 06:45:49 pm
Mr Schewe advocates keeping the photo at native pixel dimensions for printing, as long as this results in a print between 180 and 360 dpi.

Rumour has it that that fiercely defended position is no longer as rigid, since LR3 and as apparently disclosed in a recent LuLa video DVD. But then people say the darndest things, don't they?

Cheers,
Bart
Title: Re: Why upres images for printing?
Post by: feppe on September 16, 2010, 07:07:49 pm
Rumour has it that that fiercely defended position is no longer as rigid, since LR3 and as apparently disclosed in a recent LuLa video DVD. But then people say the darndest things, don't they?

Indeed, just checked the LL LR3 tutorial, he does recommend uprezzing by 50% if you're nearer to 180 dpi. There were apparently some improvements in the up/down rezzing algorithm in LR3.

He and Michael are very much advocates of doing your own testing, but their guidelines are good thumb rules from gentlemen who do know what they're talking about.
Title: Re: Why upres images for printing?
Post by: Bart_van_der_Wolf on September 17, 2010, 05:21:58 am
Indeed, just checked the LL LR3 tutorial, he does recommend uprezzing by 50% if you're nearer to 180 dpi. There were apparently some improvements in the up/down rezzing algorithm in LR3.

Some of these improvements are no doubt that there finally is sharpening added after uprezzing, as I've been advising all the time. One of the great benefits of resampling to the printer's native resolution is that one can sharpen at the highest meaningful level of detail. The sharpening can compensate for the resampling loss of contrast, and one can precompensate for the losses of the printing process itself. It can even be done at a level of detail that exceeds visual acuity, so one optionally can even exaggerate a bit without noticing the artifacts (make the image look sharper than it actually is).

Of course Qimage has allowed doing this for ages already, and it additionally allows the user to tweak the amount of sharpening to one's liking.

Cheers,
Bart
Title: Re: Why upres images for printing?
Post by: JohnTodd on September 17, 2010, 01:29:20 pm
Thanks to all for your responses. I suppose that, as an IT geek, I have trouble believing that Bart's 'highest meaningful level of detail' is anything other than the original camera pixels, in that I don't see up-rezzing adding *meaningful* detail. But feppe's point about doing one's own testing is well made.
Title: Re: Why upres images for printing?
Post by: feppe on September 17, 2010, 02:09:28 pm
Thanks to all for your responses. I suppose that, as an IT geek, I have trouble believing that Bart's 'highest meaningful level of detail' is anything other than the original camera pixels, in that I don't see up-rezzing adding *meaningful* detail. But feppe's point about doing one's own testing is well made.

One note about doing your own testing: do it at the distances you and others plan on viewing the prints. There's no point in sticking your nose to a 20x30" print if you hang it on a wall over a couch.

Then again there are people who explain in all seriousness how they do print tests with a microscope. I still think those are said in jest, but one wonders...
Title: Re: Why upres images for printing?
Post by: Bart_van_der_Wolf on September 17, 2010, 02:39:20 pm
Thanks to all for your responses. I suppose that, as an IT geek, I have trouble believing that Bart's 'highest meaningful level of detail' is anything other than the original camera pixels, in that I don't see up-rezzing adding *meaningful* detail. But feppe's point about doing one's own testing is well made.

Hi John,

Indeed doing one's own testing is alway useful, assuming one knows what to test.

A good 'upsampling' algorithm does invent data. Some of that data can be close to what it would have been if one had a larger number of pixels to begin with. A straight edge e.g. should remain a straight edge and a smooth gradient should remain smooth. A poor algorithm will introduce visible artifacts like jaggy edges or posterized gradients, there's little joy in sharpening that.

Now, starting with quality resampling (http://www.luminous-landscape.com/forum/index.php?topic=40809.msg340120#msg340120), combined with good restoration which (to a degree) corrects the blur (i.e. adds resolution) from the interpolation, sharpening at the printer's native resolution will even allow to compensate a bit for the lack of real resolution, because halo at 600 or 720 PPI cannot be seen with the naked eye, but since it boosts the gradient of edge transitions, it does make a print look sharper.

Cheers,
Bart
Title: Re: Why upres images for printing?
Post by: Schewe on September 21, 2010, 07:19:47 pm
Mr Schewe advocates keeping the photo at native pixel dimensions for printing, as long as this results in a print between 180 and 360 dpi.

Actually, that was 180-480 PPI, not 360 PPI. And that was from Bruce Fraser back when we were testing for upsampling and sharpening. In point of fact, if your native resolution at the print size is over 480 PPI, I would amend what Bruce had stated that with today's printers and larger file sizes, using the native resolution still holds but not the 480 PPI cap.

And...depending on the image type and printer, there may indeed be benefit to upsampling and then sharpening. Lightroom 3.0 had the max rez cap moved up from 480 PPI to 720 PPI mainly because testing showed that high frequency images or images with strong contrast diagonals or circles might benefit from upsampling and then sharpening. For Epson printers that number is 720 PPI. For HP and Canon printers the number is 600 PPI. This doesn't hold true for ALL images...which is why I suggest testing it out on your own images. Also note that this holds true for Lightroom 3.x printing because of the adaptive upsampling that Lightroom does. Things are a bit different in Photoshop where only Bicubic Smoother is the preferred choice for upsampling. Also note that the output sharpening done in Lightroom 3.x is designed to work for optimal results when connected to properly sharpened images in the Develop module.
Title: Re: Why upres images for printing?
Post by: feppe on September 21, 2010, 07:27:36 pm
Actually, that was 180-480 PPI, not 360 PPI. And that was from Bruce Fraser back when we were testing for upsampling and sharpening. In point of fact, if your native resolution at the print size is over 480 PPI, I would amend what Bruce had stated that with today's printers and larger file sizes, using the native resolution still holds but not the 480 PPI cap.

And...depending on the image type and printer, there may indeed be benefit to upsampling and then sharpening. Lightroom 3.0 had the max rez cap moved up from 480 PPI to 720 PPI mainly because testing showed that high frequency images or images with strong contrast diagonals or circles might benefit from upsampling and then sharpening. For Epson printers that number is 720 PPI. For HP and Canon printers the number is 600 PPI. This doesn't hold true for ALL images...which is why I suggest testing it out on your own images. Also note that this holds true for Lightroom 3.x printing because of the adaptive upsampling that Lightroom does. Things are a bit different in Photoshop where only Bicubic Smoother is the preferred choice for upsampling. Also note that the output sharpening done in Lightroom 3.x is designed to work for optimal results when connected to properly sharpened images in the Develop module.

Thanks for the correction and detail!

Getting sharpening (and noise reduction) "right" when printing is probably the toughest part after one gets color management in order.
Title: Re: Why upres images for printing?
Post by: Bart_van_der_Wolf on September 27, 2010, 08:23:41 am
And...depending on the image type and printer, there may indeed be benefit to upsampling and then sharpening. Lightroom 3.0 had the max rez cap moved up from 480 PPI to 720 PPI mainly because testing showed that high frequency images or images with strong contrast diagonals or circles might benefit from upsampling and then sharpening. For Epson printers that number is 720 PPI. For HP and Canon printers the number is 600 PPI. This doesn't hold true for ALL images...which is why I suggest testing it out on your own images.

Hi Jeff,

Just to make sure, are you suggesting that sharpening may hurt images without high spatial frequencies, or merely that it wouldn't help them? In the latter case, when it doesn't hurt, the only benefit of not sharpening at native output size would be a slightly faster workflow (sharpening a smaller size and then upsampling is slightly faster than first upsampling and then sharpening).

It seems a bit like a needless complification to me, having to decide when not to do things in an optimal way.

Cheers,
Bart
Title: Re: Why upres images for printing?
Post by: Schewe on September 27, 2010, 02:55:05 pm
Just to make sure, are you suggesting that sharpening may hurt images without high spatial frequencies, or merely that it wouldn't help them?

In my experience, upsampling and then sharpening really only helped high frequency images. It didn't "hurt" lower frequency images, but the benefits are less if at all. So, I suggest people test this on their own images and see if they can see the benefits. If yes, you can apply the upsample/output sharpening to all images.
Title: Re: Why upres images for printing?
Post by: Bart_van_der_Wolf on September 27, 2010, 03:04:52 pm
In my experience, upsampling and then sharpening really only helped high frequency images. It didn't "hurt" lower frequency images, but the benefits are less if at all. So, I suggest people test this on their own images and see if they can see the benefits. If yes, you can apply the upsample/output sharpening to all images.

Thanks for confirming. IMHO there are no ill effects to be expected, only positive ones (if there is detail to be sharpened in the first place). That's why I've always advocated this workflow for digital capture.

Cheers,
Bart
Title: Re: Why upres images for printing?
Post by: Graystar on October 20, 2010, 08:02:24 am
Not wanting to hi-jack the recent thread about *how* to up-res befor printing, I would like to know *why*.

My thinking is this. I shot an image with a certain actual resolution, cropped it and edited the remaining 'camera' pixels. If I send those to the printer, the printer driver is going to map those pixels to actual nozzle-squirts as effectively as it knows how.

What is to be gained by increasing the resolution of the image in Photoshop (or similar) rather than letting the printer driver have maximum control over the mapping? Can Photoshop create a better interpolated pixel than the printer driver?

The reason for increasing resolution is so that print sharpening can be performed.  We print-sharpen to restore the detail that is lost through the printing process.

Before I dive into it let me give an example.  Here are images of two prints made from the same image file.  The first was printed after some light and typical processing (contrast, sharpening, etc.)  That processed images was then print sharpened to give the second print.  The different is dramatic...

(http://home.roadrunner.com/~graystar/printsharpening01.jpg)

(http://home.roadrunner.com/~graystar/printsharpening02.jpg)

I can tell you from having the image and actual memory board that the second print is much closer to the actual object.  So what’s going on?  Well, printers are not monitors.  They don’t create images in the same way.  As such, the same pattern of pixels will give different results.

On a monitor, every pixel element is square and has a black border around it (take a look with a magnifying glass.)  This provides separation and definition (many computer icons will have black borders for the same reason.)  On a printer, however, every image pixel ends up as a matrix of dots, and those dots overlap one another...both within the matrix and along the border with other pixel matrices.  Dots must overlap because they are round, and round dots placed next to each other will leave white space between the dots.  This overlapping of dots has the effect of reducing the sharpness of the image.

The solution to this degradation of image quality is to create tiny halos around detail.  When printed, these halos will blend, reducing the sharpness, and providing an image that contains all the detail that can be seen on the monitor.

Before print sharpening, an image should be upsized to the native resolution of the printer (600 PPI for Canon, and 720 PPI for Epson.)  Why?  Because you want to create halos as small as possible.  At 600 PPI, a 1-pixel wide halo is only 1/600 of an inch wide.  That’s what you want.  You’re not going to see these halos...you’re only going to see the combined effect of the halos.  If your image is already going to print at 600 PPI or slightly greater without upsizing, then just leave it. 

So back to the examples.  The calculated PPI on these prints was 712 PPI, so I didn’t have to resize the image for print sharpening.  As I said, the second print was processed for printing. In this case I simply applied unsharp-mask at values of 500, 0.8, 2 (threshold should be adjusted if there are smooth gradients, like sky.) This created small-scale halos around detail.  When the image was printed, the halos disappeared due to overlapping ink dots between pixels, leaving a much improved image.  The improvement is clearly visible at a viewing distance of 12 inches.
 
There is software that will do this for you.  Qimage does a great job of restoring detail, Nik Sharpener Pro provides an output sharpening function, as does other software.  So if you don’t want to do it yourself then you can have software do it for you.  But it really should be done.  I encourage people to download the evaluation version of Qimage just to get a print and see the difference.  I was quite surprised and a bit perplexed at how big the difference was.  But now that I know why, it makes perfect sense. 

Title: Re: Why upres images for printing?
Post by: hjulenissen on October 20, 2010, 08:38:59 am
...
A good 'upsampling' algorithm does invent data. Some of that data can be close to what it would have been if one had a larger number of pixels to begin with. A straight edge e.g. should remain a straight edge and a smooth gradient should remain smooth. A poor algorithm will introduce visible artifacts like jaggy edges or posterized gradients, there's little joy in sharpening that.
...
I have an issue with "inventing data" or "creating information". I see image scaling as a much simpler process: changing the data representation grid. No new information is created, and often none is removed.

No scaling can invent new information that was not in the image. But it can make the upscaled image "more pleasing to the eye" by various assumptions about what is pleasing to the eye, and what typical objects tend to look like. But by doing this, there is a danger that what is an improvement in 95% of the cases, turns out to be a negative in 5% of the cases.

The argument for sharpening is that our eye basically use edge-sharpness as a measure for a lot of things, including perception of focus and level of details. Therefore it makes sense to have (large-scale) edges appear really sharp in the final representation grid (print).

It seems that printers do have a "native" solution where one maps most directly to squirt patterns. It seems also that optimal resolution for sharpness may be the optimal one for gradients??

-h
Title: Re: Why upres images for printing?
Post by: Policar on October 21, 2010, 11:40:59 pm
As you can see above, sharpening has a powerful effect on perceived sharpness.  But unsharp masks existed before digital capture and sharpening (subtracting a blurred negative image) is different from upressing (interpolating resolution).

In my opinion, upressing exists to emulate how film prints look when enlarged.  No, not extremely grainy, but with a gradual fall-off into a loss of detail and with no pixellation.

If you look at the modulation transfer function of a digital sensor versus film, the digital sensor cuts off from a >100% response to 0% very suddenly, whereas film transitions from >100% to 100% to 20% (beyond which it's effectively nothing) slowly and nicely.  Digital has more detail over a given sensor size and you can enlarge it way bigger, but at a certain point you see pixels with digital. Sharpening can make pixels even more apparent.

Enlarge an extremely sharp image by 800% in Photoshop using nearest neighbor and then using bicubic softer.  Pretty huge difference.

Granted, most printers probably have at least bilinear upscaling, but, like, if you upscale yourself you know what you're getting in advance, and you have freedom to choose when to sharpen, how much to sharpen, etc.  At first upressing mostly means avoiding pixellation, beyond that point it's just a desire for control in terms of sharpening, acutance, etc.  And that does matter--how much is personal opinion.  I think it might just be that some people like to see exactly what they're getting, or at least the best approximation of it before printing.
Title: Re: Why upres images for printing?
Post by: Graystar on October 22, 2010, 01:50:32 am
As you can see above, sharpening has a powerful effect on perceived sharpness.  But unsharp masks existed before digital capture and sharpening (subtracting a blurred negative image) is different from upressing (interpolating resolution).

In my opinion, upressing exists to emulate how film prints look when enlarged.  No, not extremely grainy, but with a gradual fall-off into a loss of detail and with no pixellation.

I think you’re missing the point of my post.  Unless your image is printing at a very low resolution of 100 PPI, upsizing alone doesn't do anything for you.  A print of an upsized image looks exactly like a print of the original image.  The only reason to upsize is to sharpen for print.  An image that’s been sharpened for print looks horribly oversharpened on the computer monitor.  But that’s what’s necessary when you need to view the same image on two completely different reproduction technologies.

An image that isn't processed for printing will contain less apparent detail than is available.  Upsizing and sharpening restores the lost detail.
Title: Re: Why upres images for printing?
Post by: Policar on October 22, 2010, 02:11:47 am
A print of an upsized image looks exactly like a print of the original image.

Only if you're upressing beyond the resolution a printer provides will it look exactly the same.  Otherwise, whatever algorithm you use will interpolate detail and so improve smoothness/"digital tonality"--the difference between bicubic and nearest neighbor upscaling, for instance.  Of course, most printers should resample intelligently, but not as well as photoshop does.

The only reason to upsize is to sharpen for print.

True, it's important to sharpen for a given printer and upscaling to a standard resolution facilitates that.  But there are other reasons to upscale than to facilitate proper sharpening.  Upressing fills in the gaps intelligently, giving a smooth, soft tonality when viewed up close, instead of pixellation.  (This is also why sharpening first looks so much worse--you're interpolating fake detail, haloing, aliasing, etc.)
Title: Re: Why upres images for printing?
Post by: Graystar on October 22, 2010, 06:00:00 am
Only if you're upressing beyond the resolution a printer provides will it look exactly the same.  Otherwise, whatever algorithm you use will interpolate detail and so improve smoothness/"digital tonality"--the difference between bicubic and nearest neighbor upscaling, for instance.  Of course, most printers should resample intelligently, but not as well as photoshop does.

True, it's important to sharpen for a given printer and upscaling to a standard resolution facilitates that.  But there are other reasons to upscale than to facilitate proper sharpening.  Upressing fills in the gaps intelligently, giving a smooth, soft tonality when viewed up close, instead of pixellation.  (This is also why sharpening first looks so much worse--you're interpolating fake detail, haloing, aliasing, etc.)

Do you have some before/after scans or images of prints that exhibit the quality improvements you've outlined?
Title: Re: Why upres images for printing?
Post by: hjulenissen on October 22, 2010, 09:35:51 am
...there are other reasons to upscale than to facilitate proper sharpening.  Upressing fills in the gaps intelligently, giving a smooth, soft tonality when viewed up close, instead of pixellation.  (This is also why sharpening first looks so much worse--you're interpolating fake detail, haloing, aliasing, etc.)
What you are describing sounds like regular interpolation using different filter kernels. If inserted into that framework, there is a lot of math and empiry to help us out.

An "allpass" filter would equate nearest neighbor. A triangular filter would equate bilinear interpolation. Etc. The point is that the response of such functions can be predicted from their frequency response - deviation from a flat passband and attenuation of aliasing in the stop-band.

http://www.worldserver.com/turk/computergraphics/ResamplingFilters.pdf
Turkowski wrote a nice whitepaper for Apple on this in 1990.

I am aware that some exotic image scaling algorithms does not fit into this framework, typically being non-linear, content-adaptive etc. But those are exceptions.

-h
Title: Re: Why upres images for printing?
Post by: Bart_van_der_Wolf on October 22, 2010, 10:18:38 am
I am aware that some exotic image scaling algorithms does not fit into this framework, typically being non-linear, content-adaptive etc. But those are exceptions.

Are they exceptions? They probably are in non-specific workflows, but we are talking about preparing for print output here. Here all 'tricks' are allowed (e.g. adding noise after upsampling, and enhancing edges while avoiding stairstepping), as long as the output quality 'improves' for a given output magnification and viewing distance(s).

There is a number of commercial solutions available, such as Qimage, Photozoom Pro, Blow-up, and perhaps Lightroom. There are even more commercial programs available, but their output quality varies (e.g. Genuine Fractals). Not all that exotic afterall, I would say, but perhaps you were thinking of other proprietary solutions.

Cheers,
Bart
Title: Re: Why upres images for printing?
Post by: Policar on October 22, 2010, 05:15:47 pm
Do you have some before/after scans or images of prints that exhibit the quality improvements you've outlined?

Nope!  I always sharpen after scaling so I haven't eliminated that variable.

If upscaling matters at all beyond giving you per-pixel flexibility with sharpening, adding grain, etc. it's not much.  But that's only because any decent printer that prints really large has built-in upressing software.

Your point about sharpening is spot on:  sharpening for the print helps make up for the printer's softness, not just softness inherent to the file, whereas sharpening at capture resolution (usually) only increases pixellation because you're pushing the frequency response way past 100% right before it cuts off to 0%, resulting in a very "digital" image without that soft "film-like" response I described...

I think you may be reading way more into my comment than you should be...  I basically just said, issues of sharpening aside, bicubic looks better than nearest neighbor...  If you want to see a print that proves that I'll go ahead and make one.
Title: Re: Why upres images for printing?
Post by: hacimd on November 09, 2010, 09:29:33 am
Does anyone have experience using Noise Ninja or Photo Kit Sharpener?  I've heard some good things about Photo Kit Sharpener, but not too much about Noise Ninja.  Would you recommend using either  of these if you are upscaling for large prints on a 9800?  Many Thanks!