I expect that most of the Forum readership is now beyond bored with this epic thread, but for the sake of completeness here are the latest results from Smith Labs. If you print from Lightroom to an Epson printer, they could be of interest. I have not solved the mystery of whether the OS or the Epson print driver resamples incoming data or not, but I have found out a bit more about what LR is doing. As before, my output was from Lightroom 3.4.1, running on Win 7 64-bit, and printing to my Epson R2400. The 2400 was set to its best output resolution, Photo RPM (5760 x 1440), High Speed off, and the paper was Epson Premium Glossy. LR Output Sharpening was off. My test files were 39MP Hasselblad 3FR raw files, and the specific file I used for this test is attached at the bottom of the post just for reference – this VW Campers pic is an old warhorse, but I use it because it is of good quality, well-focused and has no camera shake. I used just a section of the picture from the left of the frame, the cups on the table.
I decided to print my test only at exact divisions of the maximum 720 ppi input for the Epson, so I used 180, 360 and 720 ppi, in order to eliminate any re-sampling artefacts which might be introduced by the OS or the Epson Driver (if they do in fact resample). I started out by cropping a section of the image and sized it in the LR print module to exactly 360 ppi, fitting nicely on one third of an A5 sheet (so that I could print three tests side-by-side for comparison). This represents a section from a 20x15 print of the full-frame image. I printed it first to the 2400 directly at 360, and then tried resampling it to 720 ppi. The second result was obviously better even to the naked eye, with crisper definition and more detail on the pattern on the cups, crisper text on the tablecloth and an overall better internal contrast, particularly in the half-shadow areas. But why? No extra detail has actually been added, of course – the detail in the file is exactly the same, but you would swear it had been enhanced. At first I thought that perhaps the uprezzing routine was invoking an enhanced dither pattern in the print driver, but the dot density seemed the same under the 8x loupe. So I decided to go BIG, in order to get a closer look at what was going on.
This time I cropped a section of just the cups to 180 ppi and ran the test again – firstly a straight output at 180, then upsampled to 360 and also 720 ppi. This time I could see that LR is in fact not just resampling but running a very sophisticated anti-aliasing and smoothing routine on the image data into the bargain. It’s good at 360, and actually slightly better again at 720. And that’s why the “Schewe Rule” works. Have a look at the two files attached below – these are scans from the test prints, a section of the saucer rim on the right-hand cup. The first is the straight print at 180 ppi, the second has been resampled by LR to 720 ppi. Both are the same size in print, of course. You should be able to see how LR has smoothed the saucer rim.
Conclusions? Or perhaps a summary for Mike . . .
• We are talking about striving for the very finest prints here, squeezing the last little drop of quality out. Many people would never notice the difference – the current printers are so good, that almost any settings produce very acceptable output.
• For the best quality, set your print driver to either 5760x1440 (for the desktops) or 2880x1440 (for the Stylus Pro range). On the SP printers checking the Finest Detail box to on
may give you a touch more definition in grass, hair, and fine lines. Make sure High-Speed or Bi-Directional printing is off
• If your image sizes to less than 180 ppi in the LR Print Module for your selected paper size, you are trying to print it too large – but you may still get a reasonable result. A purpose-built uprezzing package might do better.
• From 180 ppi upwards to (presumably) 719 ppi, you will get the very best results by resampling the file in LR to 720 ppi. Between 180 and 359, you can resample to 360 to save time and spool file size, and get almost as good a result, but 720 is the business. What happens is that of course no extra detail can be added, but LR not only resamples but uses a very clever anti-aliasing and smoothing routine on the image, which enhances the apparent detail and resolution.
• This does not happen when LR is downsampling (as from 741 to 720 ppi). The image is not smoothed and no anti-aliasing is applied, so you should avoid this situation. The print quality suffers as a result (although it takes a loupe to see it). Unfortunately, with 60MP and now 80MP backs (and probably 100MP+ before too long) this is going to be of increasing concern for small prints.
• Strangely, the absolute worst thing you can do is to resample the image in LR to the same
value as reported for the cell size – for example, take a 360 ppi image resolution and then set the LR output resolution to 360 as well. Theoretically, it should make no difference – but the quality of the print is degraded compared with simply sending it to the printer with no LR resampling. I can’t explain why this happens, but it is observable and repeatable. I can also see the same result with the Restest file – comparing output direct at 720 with the same file output with the LR Resolution box also set to 720 ppi (the sample text is degraded).
• Setting the LR output sharpening to “Standard” with the correct paper type will give the print a little bit of “snap” without introducing any artefacts or halos. It’s not a huge difference with my 3FR files, just nice and subtle.
That’s my conclusions for tonight, at any rate. You are now invited to weigh in and prove me wrong