print vs LR resolution

[enduser]

I'm always encouraged when discussions get to this level.  It lets me know that this Forum has real depth of expertise. Please continue.

[hjulenissen]

Yes, and with adequate contrast. But that leads to stating the obvious, that nothing beats real resolution, and that lacking that we need proper re-sampling, and proper sharpening.

Bumping real resolution is a "trivial fix" to the problem, but does not seem to really answer the OPs question:

...How can you calculate for a given size of a  print, what amount of PPI and DPI should be chosen for having the best possible fine art print?...

Rather, the question seems to be "how low can you go on resolution while still having no visual loss" (I am assuming a properly chosen angular resolution metric). An often quoted figure is 1 minute of arc, something that can be related to the elements of letters used for establishing 20/20 vision:

http://www.1800myeyedoc.com/faqaboutus/view.nhtml?profile

A common counter-argument is that Vernier acuity extends the limit to 0.13 arc minutes (for special case images). If we need to satisfy this requirement, the "maximum print size" of those D800 images is significantly reduced. I believe that we can supply our vision with this kind of "spatial alignement of lines" without the accompanying increase in resolution requirements, thus relaxing the demands on expensive equipment or limited print size.
http://en.wikipedia.org/wiki/Visual_acuity#Other_measures

I believe that you cannot have both high contrast and linear-filter-like degradation in a sampling system as you approach fs/2. The anti-OLPF, 1:1 pixel viewing afficionados seems to prioritize high contrast, while others prioritize linear behaviour. If we are to discuss what bandwidth is "necessary", we have to also discuss (like you seem to allude to) the properties of capture, processing and rendering beyond mere passband width.

-h

[Bart_van_der_Wolf]

Bumping real resolution is a "trivial fix" to the problem, but does not seem to really answer the OPs question: Rather, the question seems to be "how low can you go on resolution while still having no visual loss" (I am assuming a properly chosen angular resolution metric).

Hi,

Well, I don't want to repeat myself too often, but I've posted about a tool to assist in finding that answer, and more... However, Vernier resolution is not a specific parameter, although one has the possibility to choose different visual angular acuity resolutions (including the practical limit of visual resolution, the "Very high quality" setting at section 1.1). There is even a theoretical Fovea cone resolution choice available, "Extremely high quality".

Of course, the actual situation is complicated by the Contrast sensitivity of the human eye (brain) which peaks around 6-8 cycles/degree.

So what's actually needed to answer the question is a complete system MTF calculation, from scene contrast, to capture device, to post-processing, to viewing distance, to output medium MTF and surface structure. Let's also not forget viewing conditions. This is not practical - if at all possible - in the majority of situations, so an approximation may be good enough.

A common counter-argument is that Vernier acuity extends the limit to 0.13 arc minutes (for special case images).

Without adequate micro-contrast there is also no Vernier resolution possible. Hence the somewhat more practical limits used in my tool. Remember that as the level of detail increases and approaches the limiting resolution of the capture process, the MTF approaches a zero response.

I believe that you cannot have both high contrast and linear-filter-like degradation in a sampling system as you approach fs/2.

Exactly, the MTF response will become too low to allow contrast detection by eye, even before we cannot measure it any longer.

The anti-OLPF, 1:1 pixel viewing afficionados seems to prioritize high contrast, while others prioritize linear behaviour. If we are to discuss what bandwidth is "necessary", we have to also discuss (like you seem to allude to) the properties of capture, processing and rendering beyond mere passband width.

Well, not using an Optical Low Pass Filter (OLPF) will boost the MTF for all spatial frequencies. Unfortunately it will also mix aliasing into those spatial frequencies if small enough detail of adequate contrast is present in the subject we're shooting. In the light of all factors, I believe there is little practical benefit beyond 0.4 cycles/mm resolution. That's why I chose that as the value to calculate with for the "Very high quality" setting of my planning tool.

Cheers,
Bart

[Ernst Dinkla]

Hi,

posted about a tool

So what's actually needed to answer the question is a complete system MTF calculation, from scene contrast, to capture device, to post-processing, to viewing distance, to output medium MTF and surface structure.
Cheers,
Bart

Bart,


That could become a nice Android App.

Pity that no output media MTFs are published.


--
Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
July 2013, 500+ inkjet media white spectral plots.

[Ernst Dinkla]

In terms of the difference between image resolution and printer resolution, the output resolution of printers refers to droplets/inch. So, at 2880 the Epson printers put 2880 droplets of ink/inch. In terms of dots/inch, that's a different measurement–printers report their resolution to the print pipeline as dots/inch (DPI) which is a different measurement unit...Epson printer report 360 DPI which Canon HP report 300 DPI. There are driver modes that can change the reported resolution; Finest Detail makes Epson printers report 720 DPI while Canon & HP can be set to report 600 DPI.

Confused yet? Go back and reread the section in the book...

Jeff,

The reported resolution should be in PPI but most drivers do not use that either. The "Maximum DPI" print quality setting in the HP B9180 driver asks for 1200 PPI input. I guess that some desktop Canons do the same. Even the old driver/firmware of the HP Z3100 had a print quality setting that asked for 1200 PPI input. It dropped to 600 PPI with the latest driver/firmware version. The humble HP K5400 Officejet Pro asks for 1200 PPI input with the highest print quality setting on photo paper. Probably a bit optimistic for that machine.


--
Met vriendelijke groet, Ernst

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
July 2013, 500+ inkjet media white spectral plots.

[Bart_van_der_Wolf]

That could become a nice Android App.

Pity that no output media MTFs are published.

Hi Ernst,

I know. If only my resources were unlimited ...

Cheers,
Bart

[bjanes]

In terms of the resolution you "need" for a print, that depends on viewing distance because human vision resolution is dependent on the viewing distance. The close you hold a print to your eyes, the more resolution that your eyes can resolve. There's a chart on page 129 that tells you the eye's resolvable resolution. A couple of examples: viewing distance of 8 inches would require 428 PPI in PPI resolution. From 18 inches you would need 191 PPI.

Confused? Yes! The chart is on page 60 of the first edition, where the resolution is in terms of pixels/inch. In the 2nd Ed, the chart is on page 76 and the resolution is in terms of dots/inch. Perhaps the change was to deal with the differences between contone printing (e.g. Fuji Frontier) and error diffusion printing as used in our inkjet printers. Human visual acuity is quoted as about 1 minute of arc or 30 cycles/degree; however, the contrast sensitivity function peaks at 6 cycles/degree. For practical viewing, we don't really need 30 cycles/degree.

In terms of determining the usable range of output resolution, Bruce thought that depending on print size, you needed at least 180 PPI to a max of 480 PPI. However, I've found that depending on the printer, you can tell the difference up to about 720 PPI (assuming Epson, or 600 PPI for Canon or HP). This type of resolution is really only useful for small prints–which is handy because you'll have higher output resolution when making small sized prints.

Now we are back to pixels/inch from dots/inch. As stated on page 83 of the second edition, with inkjet printers using error diffusion, the relationship between dpi and ppi resolution is fairly indirect, but the effective resolution of Epson inkjets is 360 ppi and that of HP inkjets is 300ppi.

In terms of the difference between image resolution and printer resolution, the output resolution of printers refers to droplets/inch. So, at 2880 the Epson printers put 2880 droplets of ink/inch. In terms of dots/inch, that's a different measurement–printers report their resolution to the print pipeline as dots/inch (DPI) which is a different measurement unit...Epson printer report 360 DPI which Canon HP report 300 DPI. There are driver modes that can change the reported resolution; Finest Detail makes Epson printers report 720 DPI while Canon & HP can be set to report 600 DPI.

For 360 DPI, what resolution in terms of PPI is needed in our image? If we were resizing manually in Photoshop, what resolution in terms of PPI would be needed?

Regards,

Bill

[Schewe]

For 360 DPI, what resolution in terms of PPI is needed in our image? If we were resizing manually in Photoshop, what resolution in terms of PPI would be needed?

I have not done any testing to determine what the least amount of resolution would be required...back when Bruce and I did PhotoKit Sharpener, we determined that 180 PPI would be the minimum prior to upsampling. But at the time Bruce didn't see any reason to upsample native resolution. At this point (and depending on the media and printer) if the native rez at print size is under 360 (or 300 for Canon/HP) upsample native rez to 360 then output sharpen. If the native rez is above 360, upsample to 720 (or 600 for Canon/HP) and then sharpen.

Yes, I've seen some lower rez output look really nice but often simply because the image itself was really great and the lack of resolution didn't matter to the image. Michael did a shot where he cropped way, way into some people walking on a road in Madagascar. The resulting image after the crop was way to small yet the image looked cool in part because of a lack of resolution.

My approach is to get the best image detail I can get then worry about the image :~)

[digitaldog]

I have not done any testing to determine what the least amount of resolution would be required...back when Bruce and I did PhotoKit Sharpener, we determined that 180 PPI would be the minimum prior to upsampling.

I've done some testing and I think 180 PPI is probably bare minimum. I recall my first high quality contone desktop printer. It was a Kodak XL7700 dyb sub circa 1993 or so. Output resolution was 203 DPI (you sent it 203 PPI files). At the time, Kodak claimed 203 was the minimum necessary resolution and keep in mind, back then, an 8x10@203 was considered a decent sized file when the best we had was a Mac IIci with 8mb of ram.

[Manoli]

For 360 DPI, what resolution in terms of PPI is needed in our image? If we were resizing manually in Photoshop, what resolution in terms of PPI would be needed?

In 2008, with the arrival of the Epson 3800, we (5 of us in the studio) carried out a rough&ready test on both pearl and glossy paper to see if we could actually observe any difference in output. We used Nikon's promo images for the D3 , exposed at 200-1600-3200 ISO, to avoid any in-house bias.

Output was onto both glossy and pearl A4 paper, at input resolutions of 360, 255 and 180 ppi.**  Photo black no matt. Printer set to 1440 ( native 360 dpi). Printed through photoshop, (no resizing through the epson driver). Both colour and B&W – not ABW.

In B&W, under close inspection , including an 8x lupe, none of us could see any discernible difference between the different resolutions.  In colour opinions differed. There was minimal difference , but arguably some felt that there was a difference in areas of continuous monochrome colour such as the blue of the baseball bats photo.

The conclusion, was that there was no visible difference above 250 dpi, and effectively 180 dpi output was identical to the higher resolutions, certainly in B&W. Today, due to the much improved resize algos and increased sensor mpx, I output at 360dpi – but if I wanted to go really large 180 dpi wouldn't worry me.

** edit - correction.

[bjanes]

In 2008, with the arrival of the Epson 3800, we (5 of us in the studio) carried out a rough&ready test on both pearl and glossy paper to see if we could actually observe any difference in output. We used Nikon's promo images for the D3 , exposed at 200-1600-3200 ISO, to avoid any in-house bias.

Output was onto both glossy and pearl A4 paper at 360, 255 and 180 dpi. Photo black no matt. Printer set to 1440 ( native 360 dpi). Printed through photoshop, (no resizing through the epson driver). Both colour and B&W – not ABW.

In B&W, under close inspection , including an 8x lupe, none of us could see any discernible difference between the different resolutions.  In colour opinions differed. There was minimal difference , but arguably some felt that there was a difference in areas of continuous monochrome colour such as the blue of the baseball bats photo.

The conclusion, was that there was no visible difference above 250 dpi, and effectively 180 dpi output was identical to the higher resolutions, certainly in B&W. Today, due to the much improved resize algos and increased sensor mpx, I output at 360dpi – but if I wanted to go really large 180 dpi wouldn't worry me.

Again, I think there is some confusion here. If I am not mistaken, you sized the images in Photoshop to 360, 255, and 180 pixels/inch and sent the images to the printer where the driver re-sized to the native resolution of the printer in dots/inch. When printing from Photoshop with With my Epson 3880, I follow Jeff's recommendations and resize to either 360 or 720 ppi, apply sharpening with Photokit, and send the file to the printer. The process is simpler when printing from Lightroom since one can specify the output resolution and sharpening parametrically without altering the master file, which is the raw file with parametric edits in many cases.

Bart might resize with an optimal method and then perform output sharpening via convolution, perhaps with a custom PSF derived from a slanted edge target. I do most of my printing from Lightroom and am satisfied with the results even though I might be leaving some image quality on the table. It would be interesting to see what Bart's approach would accomplish.

Bill

[Manoli]

Again, I think there is some confusion here …

Bill, I'm sure you're right ...

I duplicated and then resized (downscaled) the images outside photoshop to 255 and 180.  I then printed them through photoshop at their new native resolutions – but yes,  the Epson driver would have resized them to the printer's native resolution, in this case 1440 (360dpi).

For photo paper, the Epson driver gives you only  2 options 1440 or 2880. The 360 and 720 dpi are greyed out.  At the time I didn't know of Jeff, but worked closely with Ted Chau at Chau Digital UK ( DaVinci paper). Don't quote me on this, but discussing with Ted at the time, he advised me that (a) the native output was 360  and that (b) they never used the 2880 setting – they found it was laying down too much ink, for their paper,  and were having (considerable) difficulty creating their custom paper profiles – hence their 1440 standard.  

The aim of the test was to see if we could indeed see any difference in the printed output, starting out with lower resolution images – the consensus was no, we couldn't.

This was almost 5 years ago. Today, I like you use,  Lr – it is indeed much simpler and in many ways better.  But the question still remains – what is the purpose of specifying the output resolution, of say 180 dpi in Lr if the Epson driver still upscales to it's minimum native resolution – in this case 360 dpi. ?

And just in case I've missed the proverbial 'b**** obvious' – yes,  Jeff, I have bought the book, both of them, and no I haven't read 'The Digital Print' fully – yet.  But I promise to do so, soon …  

In the meantime, any pointers to any mistaken assumptions will be most appreciated !

M

[Farmer]

It's commonly believed that higher resolution (2880 versus say 1440) uses a lot more ink.  It doesn't.  You can test this by checking ink usage over a sample of prints (same image, multiple times) at the two resolutions.  The different resolutions, on Epson with variable dot size, change the option of which of the variable dots will be available for use to the printer.

[AFairley]

I duplicated and then resized (downscaled) the images outside photoshop to 255 and 180.  I then printed them through photoshop at their new native resolutions – but yes,  the Epson driver would have resized them to the printer's native resolution, in this case 1440 (360dpi).

I think the only way to really test this would be to use images shot with the same lens on the same size sensors with different photosite density.  36MP/24MP/16MP, but even then you'd have variations from the sensor microlenses and the AA filter.

[Manoli]

I think the only way to really test this would be to use images shot with the same lens on the same size sensors with different photosite density.  36MP/24MP/16MP, but even then you'd have variations from the sensor microlenses and the AA filter.

Most probably, but as I said in my original post, this was a quick 'rough & ready' test, based on only 18 A4 prints.  Nonetheless, I think the conclusion we reached is still a reasonable guideline today.  Namely, you'll get away with 180ppi for B&W and 180/250ppi for colour depending on the subject.

[Alan Klein]

How do you know how much to sharpen the file if you send it to an outside printer?

[Schewe]

How do you know how much to sharpen the file if you send it to an outside printer?

Buy PhotoKit Sharpener 2 (or do a lot of trial and error testing).

[Bart_van_der_Wolf]

How do you know how much to sharpen the file if you send it to an outside printer?

Hi Alan,

You need to know the PPI that the outside printer will use.

You can spend money on sharpening plugins, but without knowing the target PPI it remains a crapshoot. Possible exceptions are FocusMagic, which really(!) restores resolution by deconvolution, or Photozoom Pro which adds edge resolution that really helps certain images by actually adding high resolution detail.

In addition, there can be significant benefits for visual image quality if you use tools that allow to enhance certain spatial frequencies (detail size) in the image. Topaz Labs Detail comes to mind, which also uses the Intellicolor technology that leaves color unchanged despite contrast modification.

Cheers,
Bart

[chez]

I'm always encouraged when discussions get to this level.  It lets me know that this Forum has real depth of expertise. Please continue.

Yes, but expertise in what? Photography or....

[bjanes]

In terms of the difference between image resolution and printer resolution, the output resolution of printers refers to droplets/inch. So, at 2880 the Epson printers put 2880 droplets of ink/inch. In terms of dots/inch, that's a different measurement–printers report their resolution to the print pipeline as dots/inch (DPI) which is a different measurement unit...Epson printer report 360 DPI which Canon HP report 300 DPI. There are driver modes that can change the reported resolution; Finest Detail makes Epson printers report 720 DPI while Canon & HP can be set to report 600 DPI.

Confused yet? Go back and reread the section in the book...

The reported resolution should be in PPI but most drivers do not use that either. The "Maximum DPI" print quality setting in the HP B9180 driver asks for 1200 PPI input. I guess that some desktop Canons do the same. Even the old driver/firmware of the HP Z3100 had a print quality setting that asked for 1200 PPI input. It dropped to 600 PPI with the latest driver/firmware version. The humble HP K5400 Officejet Pro asks for 1200 PPI input with the highest print quality setting on photo paper. Probably a bit optimistic for that machine.

After suffering from considerable confusion, I did go back and read the books (my previous post referred to the sharpening books) where PPI and DPI seemed to be used inconsistently. On page 130 of the Digital Print, the main text states that the printer reports its resolution in terms of PPI (360 PPI for Epson and 300 for HP). In the sidebar on the same page, it is stated that with Finest Detail selected in the driver, the printer reports its resolution as 720 DPI. I surmise that PPI is the correct term here also. When printing from Photoshop one would resample to the required resolution in PPI. DPI can be specified in the print dialog. Jeff states that with Epson, DPI refers to droplets/inch, not dots/inch, but does not explain the difference.

For determining the radius for output sharpening, I presume that one would use a value appropriate for the selected PPI value. Perhaps Bart can clarify.

Bill