VIGC study on spectrophotometers reveals: instrument accuracy can be a nightmare
Update January 2009: we did get some comments on our article, see our response to the most important onces at this page.
(9 September 2008) Quality assessment in the graphic arts industry depends mostly on the use of spectrophotometers. Printers use them to check their production processes, customers use them to evaluate the print job for acceptance. Quality demands are getting more strict every year, the spectrophotometer decides whether a job is accepted, or not. But when VIGC, the Flemish Innovation Center for Graphic Communications, did a study on the accuracy of those devices, they found deviations up to a delta E of nearly 4… Which means trouble in the printing industry.
“Color quality is the biggest challenge in the printing industry.”, says Eddy Hagen, managing director and trend watcher of VIGC. “Graphic arts companies will try everything to get the colors as desired by the customer. Those customers will use it as the most important criterion to accept, or reject, a print job. Which makes the devices to measure that color quality quite essential. So you would expect that the quality of those devices is top class. But it isn’t.”
After having experienced some issues with different devices, VIGC started their first tests to compare multiple devices in the summer of 2007. “We saw some deviations between the different spectrophotometers that we use ourselves.”, explains Fons Put, senior consultant with VIGC. “So we set up a procedure to check and compare different devices. As the reference we used the GretagMacBeth NetProfiler test chart. This is a test chart which comes with a certificate stating the L*a*b*-values of the different patches, measured with three spectrophotometers under ideal conditions (updatet 14/09/2008). The certificate is valid for 12 months only, so it needs to be renewed every year. And then we measured the 13 patches on the test chart with different spectrophotometers. For two patches we also measured the repeatability of the devices, meaning 10 measurements in a row.”
Deviations up to delta E 3,77
Over the past year, VIGC has tested over 20 different devices, most of them are used by printing companies that VIGC is working with. This is in contrast with similar, smaller studies that have been done in the past. Other studies used devices that they got directly from the vendors. VIGC tested devices that are out in the field, that are used on a daily basis by companies in the industry. This gives them a very interesting overview of the capabilities of spectrophotometers in daily life. And those capabilities are not what people think… When a customer demands a maximum delta E of 2, which is often the case for quality print jobs in Belgium, he wants a device that measures the color as accurate as possible. However, the VIGC study revealed deviations up to delta E = 3,77 for specific colors. On average the deviation per instrument of all 13 patches is 1,56.
Different types, different brands
In the study VIGC encountered multiple devices of the same type or the same brand. Is there a relation between the type, the brand and the accuracy? “That’s an interesting question.”, says Put. “There was one general rule: the newer types of devices perform better. With devices that were a few years old, sometimes we got good results with the first one and bad results with the second one. Our own main spectrophotometer, which is calibrated regularly on that NetProfiler chart, was the best of them all. Another device, the same brand, the same type, more or less the same age, performed really bad.” When the measurements of all 13 patches were averaged per device VIGC found deviations from the exact value ranging between delta E 0,45 for the best device and 2,74 for the worst one. Which means that several devices showed – on average – higher deviations than the margins that customers expect from their printers for high quality print jobs. The highest deviation for individual patches was a delta E of 3,77. “Also interesting – or disturbing if you like – was that one brand had quite strong deviations in the red and orange. We found this on multiple devices of that specific brand.”
Even within a certain type of device, VIGC found very big differences. This graph shows the deviations from the absolute value for 7 devices of the same brand, the same type.
What causes the deviations?
With the older devices one major reason can be maintenance… “We know that some devices performed bad because the optics or the calibration tile were dirty.”, explains Put. Spectrophotometers need regular calibration and also periodic cleaning. Another reason can be the light source used. Put continues: “No light source has a perfect ‘spectral power distribution’. And if you don’t have much power in certain wavelengths, not that much color can be reflected in that region, which limits the accuracy of detecting small variations in that color region. An LED light source has a completely different spectral power distribution from a gas filled tungsten bulb. And both are used in spectrophotometers.”
This graph shows the difference in composition of the light source used in two different spectrophotometers.
Why not delta E 2000?
The big differences that were found can cause trouble: customers demand a delta E of 2, but their measurement device might be of a delta E of 3… A simple – and valid – solution for the industry would be to accept delta E 2000 as the formula to calculate color differences. “When people talk about delta E, they usually refer to delta E*ab, also known as delta E 1976. This is also the formula that is mentioned in the relevant ISO standards. But this formula is very inaccurate when it comes to small color differences.”, says Hagen. “I can show you a pair of colors with color difference of delta E 5 which is barely noticeable… Take a 100% and a 95% process yellow from ISOcoated. The deviation is just noticeable, but if you calculate it with delta E*ab, you get a figure of 5. Delta E* ab doesn’t really conform to the human perception of color differences. The newer delta E 2000 does. Take the same yellow color pair and you will get a delta E of approximately 1. Which conforms to the initial idea of delta E: a delta E of 1 is the smallest noticeable color difference.”
Download this testfile: how big is the color difference between the left and right? When measured this will give a very high delta E*ab, although the difference is barely visible.
When the test results of VIGC are recalculated with the newer delta E 2000, the figures become much more realistic. The overall average of all devices on the 13 patches is a rather bad 1,56 when delta E*ab is used, but a very good 0,39 when calculated with the more recent delta E 2000.
Hagen continues: “The bizarre thing however is that some experts don’t want to use delta E 2000 because it is not that good when it comes to rather large color deviations. In those cases the old delta E*ab performs better. But who is interested in the accuracy of large color deviations? I want accuracy in small color deviations. That is where the battlefield is, where print jobs get rejected. Not because the colors look very different, but because the delta E formula states that they are different… The printing industry would benefit a lot if the delta E 2000 formula would be the official formula for calculating color differences. But all relevant ISO standards only seem to know delta E*ab… Even the draft for the upcoming update (updatet 14/09/2008) of ISO 13655 on color measurement only talks about delta E*ab. Which is not in favor of the printing industry, nor their customers.”
Conclusions and recommendations
What should we learn from this study? First of all that the measurements from a spectrophotometer – at least the ones used in the graphic arts industry – is not absolute. There can be variations between different devices. Also the devices need to be calibrated on a regular basis and need to be maintained in a proper state. Periodical cleaning by the vendor may seem expensive, but what is the cost of a – perfect – print job that gets rejected due to the fact that the spectrophotometer was lacking maintenance and therefore showing a wrong figure?
Also the industry and the standard organizations need to consider using delta E 2000 as the standard to calculate color differences when judging print quality. For small color differences delta E 2000 conforms much better to human vision than delta E*ab. Rejecting jobs because of color differences should be about seeing differences, not just about measuring a certain number.
VIGC's response to some comments
Our article caused some stir in the industry. And that was our intention: we wanted to create awareness on the issues, we didn’t want to ‘condemn’ the vendors of spectrophotometers (otherwise we would have mentioned names...). Spectrophotometers are specialised tools, they need to be handled with care and with the appropriate knowledge. With care, that means: maintain them well! E.g. regular maintenance by the vendor is a cost, but it really is necessary. With the appropriate knowledge, that means: know how to use them in the right way, know how to interpret measurements and set goals that can be measured undisputedly.
In the article, a few points have been updated, based on the input we got (btw: thanks to all who delivered input!). Below is some more information on other comments.
VIGC only did one measurement, they should have taken three measurements and averaged the results
Yes, we only did one measurement and we should have done three, to have a real scientific approach. But, how many printers or print buyers do it that way? We have chosen to test spectrophotometers in the same conditions as they are used in the industry. And most printers and print buyers just take one measurement... Which can be very dangerous! Several years ago, we did a test with a ‘cold’ spectrophotometer (first use on Monday morning in winter time): it took over 10 measurements before the results were more or less consistent!
The measured devices were within ISO-specifications for spectrophotometers.
Yes, indeed. And this brings us to the goal of our article: create awareness. Many customers are demanding tolerances that are tighter than the ISO-specifications for printing (ISO 12647). With a tolerance that is often lower than the inter instrument deviation that is allowed according to ISO specifications for spectrophotometers, you will get trouble. What if the print buyer specifies a target color in CIELAB and the measured color is outside the - small – tolerance, only due to the deviation of the device used? That would mean that a job that got rejected – with either a reprint or a price reduction – for the wrong reasons! That’s why we wanted to create awareness.
A company always uses the same device, so inter instrument deviations are not that important.
Not always the case! We can’t put a percentage on it, but certainly in the case where the print buyer also has a quality department, they will check – and approve or reject – the print job with another device, maybe even with another brand of spectrophotometer. So in that scenario, where both printer and print buyer check the colors of a job, according to a specified color (e.g. a brand color), inter instrument deviations can play a significant role. And we’ve seen this in real life: printer measures the job within specs, print buyer measures it and it is outside the specs... Job got rejected.
To conclude an anecdote to illustrate why we wanted to create awareness...
A few years ago we got contacted by a printer of corrugated boxes. He was in the running for a really large order, but the customer had set quality targets and he didn’t know what to do with them... The maximum delta E that the customer had specified, was – if my memory serves me well – a delta E of 3 (since they didn’t specify which delta E, we assumed it was delta E*ab). Now consider that this was on brown corrugated boxes... The substrate itself has color differences that exceed delta E of 10! It is covered with dark spots... The customer delivered a printed ink sample. However, this was printed with a hand roller, on a nice glossy extremely white paper... and that was the color reference for printing on brown corrugated. The ink sample itself was full of lighter and darker areas, so which color was the reference??? But in the end, a certain amount of boxes should be sent to the quality department of the print buyer. And they would measure it, with their device – which was a completely different type of spectrophotometer than we are using in the printing industry - and approve or reject the print job... This is just a way to be sure to find a reason to reject jobs and demand a discount... This is not about getting the colors right anymore.
Color is something really complex and you need the right skills, the right equipment to get the colors right, to measure them in the right way.