Epson 4900 Printer Review – Addendum February 2011

March 7, 2011 ·

Mark Segal

by Mark D Segal

Introduction

Readers of the Luminous Landscape Forum (/forum/index.php?topic=49968.0and/forum/index.php?topic=49443.0) will recall that a question was raised about whether prints from the new Epson 4900 show colour inconstancy (CI), (often wrongly called “metamerism”). CI essentially means that the appearance of a unique hue changes according to differences of the various sources of illumination in which it may be viewed. A famous example from yesteryear: the Epson 2000P, where grays under incandescent light looked greenish under daylight (Metamerism  is about the colour of two objects looking the same under one illuminant but different under another.) Interest was also expressed in how the 4900’s prints compared with other contemporary Epson models such as the 9900 and the 7890.


Questions of Interest and Procedure

Responding to these questions and concerns, we are fortunate that Forum member Phil Brown (“Farmer”) was able to make test prints on three Epson printers – the 4900, the 9900 and the 7890. Icing on the cake, Phil offered to send me the samples all the way from sunny Australia to cold and blustery Toronto for a second pair of eyes to look at them and see whether my observations would differ from his. I gladly agreed, the prints came, and gave me a unique opportunity to examine them visually and quantitatively. You may agree – the harder it is to see differences between prints just by looking at them, the more work it creates for the reviewer, because others have claimed certain differences; therefore, one needs to haul out the spectrophotometer, fire-up Excel and do some measuring and calculating.

Phil used the image shown in Figure One, which I also use.

Figure 1. Printer Evaluation Image, ProPhoto (markings mine, explained below),

Courtesy: Bill Atkinson and Outback Photo (http://www.outbackprint.com/printinginsights/pi048/essay.html)

Once Phil completed printing and examining the test images, he came to this conclusion (post 44 of the 4900 Review discussion thread):

“Well, I’ve completed my tests.  4900, 7890 and 9900.  I’m not seeing any issue or any significant differences between the machines.  There are some differences between the 10 and 8 colour machines, but you have to look for it and really it’s helped by having them side by side.”

My visual examination of these prints confirms Phil’s conclusions. Therefore I did go the extra step(s) of hauling out the spectro, firing-up the applications and doing some (really, many) measurements. Phil provided prints on Epson Enhanced Matte (EMP) and Premium Luster Photo Paper (PLPP) using the profiles provided with the Epson driver, while I added a print of the same image on Ilford Gold Fibre Silk (IGFS) using the custom profile I made using XRite Pulse Elite. The rest of this article is about observations and conclusions from the numbers. But first a compound word about “CI”: <Forget-about-it>. None of Phil’s prints or mine reveal CI to my eyes.

Other key questions of interest in the light of the various discussions about the printers were:

(i)           how do they compare with respect to maximum Black and tonal separation at the very dark end of the luminosity scale,

(ii)          how do they compare reproducing trace differences of highlight detail;

(iii)         how do they compare with respect to grayscale neutrality,

(iv)         how do they compare for colour and chroma,

(v)          which print more accurately relative to the L*a*b* colour values of the evaluation source image and for all of these comparisons,

(vi)         are the more important differences due to the printers or the papers?

The main conclusion of this work is that paper differences are far more important differentiators of outcomes than performance differences between these printers. Differences could also be due to profiles, which I could not evaluate.

My quantitative procedure was as follows (all in L*a*b*, referring to Figure 1):

(a) Measure the paper white for the three papers; (question (iii)).

(b) Measure Maximum Black for the three papers from the upper right patch in the Evaluation Image shown with a white “X”; (question i).

(c)  Measure each step in the very dark and highlight step wedges outlined in red in Figure 1 (questions (i) and (ii)).

(d) Measure red, green and blue patches, and a middle gray patch as shown by the “X”s in the set of patches sitting above the grayscale gradient; (questions (iii) and (iv).

(e) Calculate chroma (C) values for the RGB patches using for each the standard formula C = SQRT[(a*^2) + (b*^2)]. (Chroma is “colourfulness” from gray”, as shown by absolute a* or b* increasing from “0”); (question (iv)).

(f)   Calculate differences of values between printers for the same paper and between papers for the same printer, and estimate visibility of differences; (question (vi)).

(g) Calculate difference between source image values and values read from the prints; (question v)).

I made the measurements with my X-Rite Pulse spectrophotometer, from which they are transcribed to the Colorshop X Scratchpad and then transferred to Excel.


 

A very quick primer in L*a*b* colour space measurement for those who may need it: L* measures luminosity on a scale of 0 (Black) to 100 (White). a* measures the Green to Magenta axis on a scale of -127 to +128. b* measures the Blue to Yellow axis on a scale of -127 to +128. When a* and b* both equal 0, the colour is gray, its luminance determined by L*. On the a* axis negative numbers tend to Green and positive numbers tend to Magenta. On the b* axis, negative numbers tend to Blue and positive numbers tend to Yellow. Chroma increases with absolute increase of numeric value. Hence, for example, b*+60 is more Yellow than b*+30, and a*-60 is more Green than a*-30. Any RGB colour in a given colour space can be defined by its L*a*b* values.  An L*a*b* model schematic (bars not to scale) is shown below. (Please also see the technical annex on L*a*b* measurements and print matching.)


The next issue is that of the size of the difference between two measured values for the difference to be noticeable in a print, and along with that, the consistency of the measuring device. Dealing with latter first (easier), Figure 2 shows that the Pulse is highly consistent between repeated measurements of the same colour patch.

Figure 2. Device Consistency

Regarding how much of a difference between measured values is noticeable, I consulted with noted colour management professional Terry Wyse, who has made some quantitative observations mainly based on offset printing. He advises:

<In the case of delta a* or b* for neutrals, we can just start seeing a difference around ±1.0 to 1.5. [ed. The delta values being discussed here are based on simple subtraction, which is unadjusted for the non-linearity of human visual perception. As such, the method is akin to a dE76 calculation[1]].

<For L*, a delta L* in the range of 1.5 to 2.0 starts to become noticeable. But much depends on how one is comparing. Two prints separated by several inches to perhaps a foot apart would take a delta of 2 L* to notice a difference; but lay two patches next to each other and you’ll probably see a difference at around 1.0 to 1.5 L*.

<When it comes to delta a* and b* for colour (i.e. non-grayscale) it gets more complicated. The easiest way to deal with delta a* and b* for colours is to break them down between their hue and chroma [ed. one being a*, the other b* depending on the colour being examined]. In terms of visual differences, a hue shift is more noticeable than a chroma shift.>

As I wanted to quantify visible differences in selected colours, as revealed by differences in their a* and b* components either between printers or between papers, or between reference image values and print values, I prepared printed patch tests narrowly around the evaluation image’s own a* and b* values for selected in-gamut colours (see below), in order to determine thresholds of visible differences when the patches are printed on Epson Premium Luster Paper. My findings are that it becomes very hard to detect differences in prints when the sampled colours had differences of less than:

Red: a* ~3 levels; b* 2 levels

Green: a* and b* 2 levels

Blue: a* 5 levels; b* ~3 levels

Bottom line here: we can tolerate some differences of measured values before we would see a difference in one printer or paper versus the other. Therefore in deciding which measured differences between the test prints are likely to be noticeable, I discounted all the grayscale and luminance differences which lie between -1 and +1; for the non-grayscale colours, I discounted red and green differences between +2 and -2 and the blue results between +4 and -4 (averaging for the difference between a* and b* thresholds). Differences are easier to detect comparing patches side by side rather than comparing whole images side by side, hence these boundaries are rather severe. This is useful to remember for the discussion of results below.

Reverting to Step (d) and Figure 1 above, three key things are being tested when comparing the outcomes between printers and papers: the printer, the paper and the profile (henceforth: P3). Characteristic differences between any of them can produce visible differences in outcomes.


[1]I do have a rather complex spreadsheet for calculating dE2000 values which are intended to account for the non-linearity of human visual perception; however, I chose to simply subtract one data set from the other and discount an observed range of unremarkable variance as discussed here. It’s easier to grasp.


 

 

Figure 3. Colour Patches (Blacked Patchers Are Out of Gamut)

Figure 3 (upper) shows all of the square colour patches in Figure 1, while the lower panel shows which of those patches are out of gamut (the ones covered in black) for IGFS paper in the 4900 printer. By measuring print performance differences for R, G and B in the top row of Figure 3, (as shown by the top row of Xs in Figures 1 and 3), we are really testing for how well P3 deals with out of gamut colours, hence all of these results will differ quite largely from their corresponding L*a*b* values in the source image, as indeed I’ll show later. To get a better gauge about how well P3 can reproduce source image L*a*b* values where there is no clipping, I also used in-gamut colours – the R and B from the second row and the G from the third row, shown by the Xs in Figures 1 and 3 (lower).

For the middle gray value, I selected the sixth gray patch from the left in the bottom row which has an L* source image value of 50.

All of the measurements and outcomes are contained in an Excel spreadsheet configured to address the key questions mentioned above. (The spreadsheet is available for personal use on request by email with your real name and email address.) There are about 840 measured data points and 840 results data points in the spreadsheet, notwithstanding the very selective sampling procedure described above. As it would be very tedious and unnecessary to describe every finding from an examination of this data, I extract the information illustrating the most important findings relevant to the purpose of this exercise.

Here’s a summary of outcomes:

(1) In comparison after comparison, it really does emerge that paper overwhelms printer every time in terms of differential impact on outcomes.

(2) On the whole, there is little to distinguish results between the 4900 and the 9900, but a bit more differentiation between these two and the 7890, particularly affecting Green, where the former have a larger gamut and closer to the reference image values. This is expected.

(3) Maximum Black is remarkably stronger for the gloss papers than for Enhanced Matte, hence those looking for the deepest blacks and greatest separation of dark shadow detail would be better to print on a gloss paper. This is also expected and well known.

(4) On the whole, IGFS performs more accurately compared with Premium Luster, because its Paper White is just about neutral, while Premium Luster has a noticeable bluish cast.

(5) On the whole, none of them are “dead-on” reproducing the in-gamut reference image numbers, but IGFS in the 4900 performed best in this respect, probably because of its near-neutrality and perhaps the custom profile. That said, the fideity of these printers is remarkable, especially in light of the considerations in the “Note on L*a*b* Measurements and Print Matching” near the end of this article.

Below I provide data illustrating these conclusions (not in order of presentation above). From panel 2 onward, yellow means the highlighted difference may be perceptible.

Re panel 1 above, the very small differences (between printers) of paper white for the same paper reflects instrument variance from one measurement to the next. There is a large difference of Maximum Black between the matte and gloss papers. (In reading the above chart, “4900 IGFS” means the Epson 4900 printer using IGFS paper. The “4900” results in the far left set relate to EMP in the upper rows and PLPP in the lower rows.)

Re panels 2 and 3 above, one can consider the output of the 4900 and 9900 printers to be largely indistinguishable, and that’s the appearance one sees on visual examination. The neutrality of the b* axis in the deep blacks is slightly less on the 7890 than the 4900 by the numbers, but it’s hardly detectable examining the prints. The visible difference criterion (±1) is perhaps overly strict for the grayscales. Both the in-gamut and out of gamut Green show less chroma in the 7890 compared with the 4900. Otherwise, the differences are quite unremarkable.

The bigger differences show up comparing different papers for the same printer:

Comparing the amount of yellow highlighted cells in this data (PLPP vs. IGFS) with that comparing the printers, it is clear that differences between papers matter more than differences between the printers compared here. In this comparison both are luster papers with similar maximum black, but paper white is visibly different.

Finally, a colour management favorite: how do the numbers measured on the prints compare with their corresponding source image values? Unsurprisingly, none of them are identical, some are closer than others, and some marvelously close.

Simply glancing at the yellow highlighted areas in Panel 5, IGFS in the 4900 emerges with the overall closest match between the source image values and the values measured from corresponding patches on the prints. This is not surprising, given the paper’s comparative neutrality and custom profile. It is, nonetheless, remarkable how small many of the value differences are, despite being highlighted in yellow, especially recalling that the threshold criteria for visibility of the differences on paper may be a bit too strict. Looking at the Chroma data, again overall, IGFS shows up as “best in class” within this set of comparisons. In-gamut Green is considerably closer regardless of the paper for the 4900 versus the 7890, consistently with observations above. The in-gamut RGB results are of course better than the out-of-gamut results because they are in gamut, whereas the first row of RGB file numbers (Figure 3) are all out of gamut and will be clipped to the printer profile gamut boundary.


Side-bar on Within-Model Variance

While Epson is said to manufacture its professional printers to a high common standard of performance, I nonetheless conducted one test to check for performance consistency between Phil’s 4900 and mine. After making a number of repeat measurements to insure against flukes, it turns out that there are several differences exceeding the threshold criteria of visibility, as shown in the chart below.

The left panel shows the differences between Phil’s 4900 and mine on PLPP both printed with the same profile in the same manner. The right panel shows the differences between my 4900 on PLPP and my 4900 on IGFS. Focusing firstly on the differences between the two 4900s in the deep blacks, despite the yellow highlighting I don’t consider the size of those differences to be of practical consequence, given where they sit on the tone scale. Recall, it is a slightly different positioning of tones on the blue to yellow spectrum in the blackest portion of the luminance scale. There are also several large-ish colour differences in the a* or b* channels of the red and blue colour patches which do result in slight perceptual distinctions when comparing the prints closely. Comparing the right and left panels, however, my general conclusion that paper differences matter more than printer differences is confirmed.


 

I conclude this little odyssey with a word of caution: the criteria for observing when numerical differences matter in a print are based on my own visual perception using contiguous patches printed on one paper (PLPP) using one printer (Epson 4900), and the sampling of colours chosen to both define the criteria and make the comparisons is also necessarily limited. A truly comprehensive scientific approach would require using more printers, more papers, more measuring instruments, more colours, perhaps more refined criteria for the dark end of the tonal scale and more observers. But, I am not a colour science laboratory and even this range of research was quite time-consuming. All that said, we are publishing this material because we believe it has considerable indicative value. If we were to measure many more colours, it is not likely that the general conclusions here would differ much, if at all.

Mark D Segal (+)
Toronto, February 2011.         

(+) Many thanks to Phil Brown, Eric Chan, Rich Wagner and Terry Wyse for their review and advice.  I, of course, am totally responsible for how I used it.


Note on L*a*b* Measurements and Print Matching

L*a*b* values are computed with a  “reference illuminant,” which in the case of printer profiles is almost always D50 (~5000 K CCT). The comparisons here are consistent in respect of the reference illuminant.

I mentioned that the a* and b* values lie in the range -128 to +127. This is true of ICC profiles, which use the 8-bit integer encoding range of [-128 to +127] as a convenience for implementation. However, the gamut of perceivable colors for a given illuminant such as D50 is not the shape of a “box” in the L*a*b* color space, but is instead an uneven shape, extending a bit outside of the {0,100,-128,+127,-128,+127} box, but also not completely filling the box. Interested readers could pursue this on Bruce Lindbloom’s site, in particular, his visualization of the L*a*b* gamut for D50: http://www.brucelindbloom.com/LabGamutDisplay.html, as well as more info about the [-128,+127] integer encoding, and its limits: http://www.brucelindbloom.com/LabGamutDisplayHelp.html. The discussion here, however, is (as usual) anchored in the traditional implementation applying to ICC profiles.

As well, we should have reasonable expectations for the potential match between L*a*b* values of the reference image file and the measured print. One reason is that the image file values do not have an intrinsic reference point for the image contrast. For example, consider a L* value of 0 for the black patch in the image test file, and a L* value of 100 for the white patch. How much brighter should the white patch be than the black patch? This is essentially unspecified.

In contrast, the measured L*a*b* values of prints correspond to reflectance, and are limited by the paper white and the deepest black that can be printed (such as L* ~= 5 for a glossy paper, and L* ~= 16 for a matte paper). Hence there is a natural remapping (for contrast, as well as out-of-gamut colors) that must take place, driven by the printer profile. With a relative colorimetric intent and black point compensation, for example, a black patch in the source (L* = 0) is typically remapped to the deepest printable black (e.g., L* ~= 16 in the case of a matte paper), while a white patch in the source is typically remapped to the lightest tone (e.g., paper white), and the neutral axis between these two extremes is remapped/compressed accordingly.

(I thank Eric Chan for these additional insights.)


 

Addendum on Print Preview

Users of the Epson Driver for Windows are accustomed to a very handy feature in the driver allowing one to select the option to automatically preview the print as it appears spooled to the printer; printing does not start until one clicks “Print” on the preview window. This preview is not colour-managed in Windows. Rather, its purpose is to allow us to verify whether the print spooled completely and the layout is correct before printing.

There isn’t such a handy feature provided as a check-box option in “Printer Settings” within the Epson driver for Mac, as it is for Windows. One can click “Preview”, but that brings up Apple’s Preview application, from which one can print, but I haven’t yet achieved clarity on whether this reliably respects the colour-managed workflow I have established between Photoshop and the Epson Driver, and I’ve also found that having selected this feature once, it remains sticky during the same session.

However, by poking around a bit I discovered that there is indeed an undocumented feature in the Epson Driver for Mac OSX allowing one to preview the image, colour-managed, just between the time it is spooled and starts printing – enough time to stop or pause the printing process if necessary. It is accessed as follows. After clicking “Print” in Photoshop, the icon for the Epson 4900 pops into the dock. Click the icon and the print job manager opens. It looks as shown on the left side of the screen-grab below. Immediately highlight the current print job and double-click it. After several agonizing moments, a full colour-managed preview of the image from within the Epson Driver (not Apple Preview) appears, as shown on the right side of the illustration. Click the little arrows under the image to maximize it on your display, or click “X” to close it, or “Stop” or “Pause” to interrupt the printing process. Unless you too have discovered this, you saw it here first!

February, 2011

The Epson Stylus Pro 4900 May be Purchased From Our Sponsor B&H Photo-Video   


 


Mark has been making photographs for the past six decades and started adopting a digital workflow in 1999 first with scanning film, then going fully digital in 2004. He has worked with a considerable range of software, equipment and techniques over the years, accumulated substantial experience as an author, educator and communicator in several fields and is a frequent contributor to the Luminous-Landscape website. Mark developed a particular interest in film scanning and authored the ebook “Scanning Workflows with SilverFast 8, SilverFast HDR, Adobe Photoshop Lightroom and Adobe Photoshop” available on the SilverFast website. In his “other life” (the one that pays for the photography), Mark is a retiree from the World Bank Group and now a consultant in electric power development.

You May Also Enjoy...

Red Window

January 13, 2009 ·

Michael Reichmann

  Please use your browser'sBACKbutton to return to the page that brought you here.


Landscape & Environment

I Dreamt I was a Butterfly By Chuck East

May 24, 2019 ·

Josh Reichmann

FacebookTweet Very rarely do I stumble upon a photographic project that I feel compelled to support because of its sheer beauty and inspiring story. Chuck