A Real-World Epson 7600 Resolution Test


There has been a lot of discussion recently about the resolution of the new Epson wide-format printers.  These discussions frequently revolve around issues such as: Can you see more detail at higher resolutions? Should I print at even fractions of the 1440/2880 resolution? What is the resolution?

I conducted a test determine a practical resolution limit for real-world color photography on the Epson Stylus Pro 7600 printer.  This test was performed using Ultrachrome inks loaded with Photo Black on Epson Premium Luster roll paper, at 1440/2880 dpi.  I will later revisit this with the Somerset Velvet for Epson paper with matte black Ultrachrome ink.  I currently have no plans to test photographic dye or 720 dpi printing.  The results should apply equally to the Epson 9600 and the larger 220ml ink carts.

Finding the resolution limit

The approach I used was one of brute force.   I started with a large-format scan and set it to 400 ppi.  Then, I cropped a 3x3" square and resample it bicubically in Photoshop 7.0.1 down to 340, 320, 300, 280, 260, and 240 ppi.  While the 400 ppi original would be printable as is (the scan is certainly well enough defined), I downsampled it to remove any possibility of scanner artifacts, or different amounts of bicubic artifacts.  In practice, I would not expect there to be any difference regardless of whether a high-quality scan or a downsampled version is used, but for this test I felt it best to eliminate extraneous variables.  The original 400 ppi crop was duplicated and resampled for each resolution to avoid bicubic step artifacts.  These crops were then transformed using a PS7 action into a test document each, consisting of 7 patches; one of the original crop at the given ppi, one given radius 0.4 50% threshold 0 unsharp mask in Photoshop, as well as one each for USM 0.4/80%/0, 0.4/125%/0, 0.5/50%/0, 0.5/80%/0, 0.5/125%/0.  Threshold 0 was used since there is no CCD or film noise to suppress.  The strips were saved as six separate documents, one for each resolution.  Part of the reason for the square patches is so I can print them rotated 90 degress side by side in the same strip.

The image chosen for the test is shown below.  The dwarf cypresses were photographed in Everglades National Park.  The photograph was made using a Nikkor SW 75/4.5 lens with a Schneider 3b center filter; 1/4s f/32.  A small amount of rear tilt was employed to nudge the perspective where I wanted it.  
The film is Fuji Provia 100F (RDP3, Quickload) scanned on an Imacon Flextight Precision-II at 1800 ppi.  The Imacon Colorflex 1.9 software was set to disable all sharpening and processing. The resulting 16-bit RGB scan was converted from the Flextight scanner profile to Adobe RGB (1998) and adjusted for levels and tone in PS7.  Adaptive light 5 was used to center the histogram in the scan (to put the bulk of the image in the color gamut sweet spot).  The scanner was focus calibrated and focus checked by scanning small crops with different amounts of descreen.  Calibration was spot on.  I chose this particular image due to its high content of fine detail such as twigs, air plants, and grass, and because of the importance of reproducing this textural detail faithfully in the print.  This particular image is carried by fine detail and its composite sense of texture.

Dwarf Cypresses, Everglades NP

The highlighted box shows the crop area selected for the test.  Below is a crop from one of the 340 ppi patches to show the level of definition (with USM 0.5/125%/0).

Crop of crop: 340 dpi, USM 0.5/125%/0

Below is a sample patch strip; it's the left portion of the 240 ppi strip.

Left portion of 240 ppi strip

The strips were printed with auto cut disabled, and the paper manually repositioned for each document.  Printing was performed through PS7's Print with Preview, using Bill Atkinson's 9600 PLU1 profile.  The printer driver was set up as per his instructions for that profile.  A curve was applied to lighten the shadows somewhat (hence the tonal disparity between the full view and the crop).  Resampling and sharpening was done in 16 bits, all other work in 8 bits (text, strip creation, printing).

Once done, I left the grid to dry overnight.  The next day I took a closer look at the patch grid.  One thing was immediately obvious: the level of detail in the image doesn't noticably affect hue or tone!  This is something I had personally been somewhat worried about, and it's comforting to know the gamut doesn't change with the spatial content.  Next, I did a cursory overview, eyeballing definition and sense of texture.  There are definitely noticable differences, though they are subtle.  On the 340 ppi strip, the 0.5/125%/0 patch looks best; on the 240 ppi strip the 0.4/125%/0 patch looks best.  They are hard to compare, though, because they're phsycally at different ends of the paper.

The next step was to take a 10x loupe to the patches to see where the resolution starts dropping off.  Easier said than done...  Frankly, I can't tell the difference.  At one moment, 300 ppi 0.4/125/0 looks best.  Then 340 ppi 0.5/125/0 looks better.  Then something else, and I'd go around in circles.  I tried to lock onto extremely fine detail in the 340 ppi patches and then look for the same in the 240 ppi patches.  Still there?  
Does it look softer, worse?  Yes... no... maybe... sigh.   Since this was getting nowhere, I held the print up next to my screen, showing patches at 200%.  Now, clearly the 340 ppi image showed some detail not in the print.  Checking the 340 ppi strip against the image data for 320 ppi, I could still find unprinted detail.  The same was true at 300, 280, 260... and that's just about where it started matching.  The 340 ppi print patches show no more detail than what's in the 240 or 260 ppi image data.  (They differ only by 10%, so the difference between the two image resolutions in terms of actual detail is rather minimal.)

I attempted to make
close-up images to illustrate my observations, but due to lack of consistent lighting (I used a flashlight for illumination) and print flatness issues, I didn't manage to make accurate dot pattern magnifications.  I'll have to think about this some more.  Meanwhile, I don't think it would be too hard for the truly suspicious reader to judge for themselves.

So what about the discussions revolving around printing at even integers, someone who heard that a friend was told by Epson to print 300 ppi, etc?   It makes no difference.  The Epson 7600/9600 UC can't resolve 300 ppi.  However, there is another aspect here, control over sharpening.  The patches do look different in terms of definition  and sense of texture.  How can this be if they contain the same detail, in fact are virtually identical, and received the same exact amount of sharpening?  E.g., 0.4/125%/0 looks nice and sharp at 300 ppi, but a little undefined at 340 ppi.  The answer is simple: the radius setting in USM is measured in pixels, and represents a different physical radius on paper when the ppi changes.  And a small change in USM is very noticable in the print, especially around the area where it goes from perceptually soft to perceptually sharp; it happens very quickly.  Where this happens depends on the contrast in the original image, the color space, the ink, the black ink used, and the paper.  It may also depend on 1440/2880 settings.

Lower resolution input for 1440 dpi prints

What about lower resolution input?  What's the difference between printing images that have detail levels of, say, 150 or 180 ppi, versus 260 ppi?  What does the output look like and is anything lost?  To answer this, I prepared additional patch strip documents for 150, 180, 200, and 220 dpi.  These were printed as previously described and placed side-by-side with the old ones for analysis.  At first glance they all look reasonably good.  Then, putting the loupe to the 150 ppi was a real shocker -- pixels!  And lots of them.  Square little things.  Clearly visible in high-contrast detail, such as a whitish branch with yellow moss against a dark brown background.  Yikes!  Apparently the Epson driver uses a near-neighbor style resampling algorithm, or at least something that approximates it for fine high-contrast detail.  The softer detail shows none of this, however.  I'm disappointed in Epson.  After discovering this with a 10x loupe I can now see the telltale signs of it by eyeball up to 200 ppi.  With a loupe I can just barely discern it at 220. Clearly this is a minimum resolution for the Epson 7600!  If you have less to start with, resample it to a higher resolution using bicubic, stair, genuine fractals, Qimage pro, VFZoom, or whatever you have on hand.  Any of these will produce a better result than the Epson driver.  Just don't use a low-resolution image as is!

Printing at 2880 dpi

The biggest question remaining at this point was what, if anything, is gained by printing at 2880.  The Epson 7600 offers an option to print at twice the resolution, but only along the horizontal axis.  Again, I printed patch strips; in the first pass I printed the 180, 240, and 300 ppi strips.  One thing was immediately obvious by just carefully inspecting the prints : 2880 dpi shows better definition.  By this I mean that edges are cleaner and color is smoother, although the latter is only visible through a loupe.  Even the 240 ppi patch looks better at 2880 dpi than at 1440.  Curiously, I couldn't find more detail in the 300 ppi patches than in the 240 ppi ones, it's just better defined.  So the same 240-280 ppi image detail level would seem to apply.  However, even though there is no more detail, 240 produces noticable jaggies while 280-300 is smooth.  Seeing this, I've decided to do my prints using 280ppi at 2880 dpi output and use 1440 for drafts and images with 240 ppi or less of image data.

Finest Detail

There is a setting in the Epson driver named "finest detail" that's not entirely clear what it does:

Printer Settings Dialog

All my previous tests reported here were performed with this setting disabled.  To determine if it adds something, I reprinted the 260 and 300 ppi strips with this box checked, at 2880 dpi.  Under a light it looks identical in terms of tone and hue, so this option doesn't appear to have any immediate effect on device calibration and profiling.  Eyeballing the 300 ppi prints closely I can perceive no difference.  Even under a loupe, they look the same.

Conclusion and recommendations

  • Make sure to print at least 220, preferably 240 ppi with 1440 dpi.  If you have less, upsize.
  • Any image detail above 260 ppi is lost in 1440 output, and I'd go so far as to say there is no difference between 240 and 260.
  • USM makes a huge difference, and lack of proofing makes it difficult to master.  Clearly this is an area that requires lots of experience.
  • USM is easier to control at higher resolutions since it's easier to make fine adjustments to the radius parameter.
  • 2880 has somewhat better definition than 1440 but you want to feed it at least 280 ppi, preferably 300.
  • If you have high-ppi image data, by all means use it.  It won't necessarily show more than 280 ppi even at 2880 dpi though.

Dot Pattern Samples

Dot pattern closeups are in a separate page.

Jan Brittenson <bson@rockgarden.net>            3/16/2003