New research results on Colour Deviations ΔE2. Oktober 2020
The technical reproduction of true colour tones, for example using printing processes, is quite difficult, because it depends on many different factors, such as thickness of ink layers (mostly measured indirectly as optical densities in the print), the data quality during the pre-print process, the kind of the substrate, being printed on, the accuracy of printing and so on. All these factors are difficult to measure and their interdependence is hard to describe. For this reason, a summarizing colour measurement is prevailing in the printing industry, which is based on the CIE L*a*b* system. Generating data for the values L, a and b, by spectroscopic measurement, certain chromaticity coordinates can be defined. Even though the CIE L*a*b* system is quite old and based on many compromises, it still is the most feasible system for colour measurements and widely used until now.
When L* (the brightness between zero and 100%), a* (the colour deviation on the red-green axis) and b* (the colour deviation on the blue-yellow axis) are measured for a master copy, the data can be compared to the chromaticity coordinates of any further printed copy, calculating the colour deviation ΔE. As ΔE just describes the average Euclidean distance between two points (original and copy) in the three-dimensional colour space, which is a random determination without deeper scientific justification, it constitutes another simplification of the reality.
The scientific problem behind, is that all benchmark data used, is based on subjective colour impressions of some test personnel from the very beginning, and also the way ΔE is calculated, is quite arbitrary. Even though some spectroscopic data is measured, it does not heal the general subjective and arbitrary input all over the process. The mathematical figure ΔE doesn´t say much about the real conditions, though, or, as computer scientists use to say – I apologize in advance – “shit in – shit out”. Ok, it is not really that bad, and, first of all, we do not have better methods, yet.
For practical use, it is not done with chromaticity measurements and calculation of ΔE, it is also necessary to define tolerances for the colour deviation, expressing which ΔE is still acceptable and which is not. As a rule, values for ΔE between 0 and 3 are considered to be sufficient (not visible at all or only visible to well versed experts), figures between 3 and 5 or 6 are seen as critical (possibly visible to some ordinary persons) and results higher than 6 are believed to be inacceptable waste (visible to everybody).
Except the quality managers, especially marketing people believe in ΔE and its importance. The reason is that they are afraid of Customers, who could not recognize their brand products by colour. People in supermarkets seldom take a closer look to the range of goods, offered, but look single-minded for products they already know. The colour of packages is an important support to find a certain product. Marketing people believe, if the colour deviation is too big, people may not find their products and thus buy different brands. Therefore, they try to put pressure on their suppliers, especially the printers, in order to reduce ΔE to an utmost minimum. Minimal allowance for colour deviations, of course, creates a lot of technical trouble and extra cost in print production. The printers therefore try to push the agreed ΔE tolerances as high as possible which may result in serious disputes with their Customers. By the way, it is always better to put any tolerance for ΔE into contracts instead of none, because otherwise you may end up in law court – with no hard facts in your hands and without any idea how the Judge will decide. There is an old saying: “before the Court and on the high seas, we are in God´s hands”. This might be not the best management principle, indeed. Thus, try to put a ΔE figure into all your Contracts, and try to make it as high as you can!
But how about the dogma of too high ΔE values and their bad influence on marketing (and other) purposes? Are our Customers really that eagle-eyed, as we tend to believe after all the theory?
A study of Beuth University of Applied Sciences Berlin has shown, that the real ΔE could be much higher than everybody believes. A large test was designed. The target was actually a different one, namely to find out if average people recognize deviations of graphical structures, such as security features. Former studies had shown, already, that people do not really mention even big deviations. For example, the picture on the label of the well-known hazelnut-nougat cream “nutella” was altered in different ways and average people were ask to tell if the picture is original or not. Although the broad majority of participants knew nutella, a vanishing minority (low single-digit percentage) saw some (! not all) of the alterations. Even when the name of the product was slightly altered (“nutela” instead of “nutella”) and even when the pictured hazelnut was replaced by a walnut, the knife by a spoon and so on, people didn´t see it. It was shattering, because it says a lot about the recognition value of holograms for security purposes, where the pictures are much more tiny and difficult to see then on a nutella label. Another experiment proved these sad results, when on a known box the main picture of another product was enlarged in steps of 25%. Even when enlarged to 200% (with the necessity to cut important parts away, to place it still on the box) some 30% of people did not mention any difference (75% enlargement was not seen by about 50% of people, plus 50% was mentioned by just one fourth of all participants, only).
In another study, the value of colour for recognition purposes was checked.
The experimental set-up was different from the normal settings for ΔE evaluation, because the participants did not have both samples available, at the same time. This is a realistic situation, because when looking for a product in a supermarket, one does not have an exact example at hand, but must remember how the required product probably looks like. If you put two samples together to find ΔE, you may see even small deviations. But this is not the reality! The reality is that you have to remember a certain product (including its colour). This was the reason for the experimental set-up.
It was not easy to create conditions for an unaltered output. People must not know what the research was all about. Most of the time was filled with pre-experiments, though.
In the end, a representative product was chosen for the test. It was CocaCola, but, of course, not in its typical bottle, to avoid people to identify it by shape, but in cans which do not differ from other cans with different content and Brand. Another problem was to erase the characteristic lettering of CocaCola, e.g. by turning it to the back side. The prepared cans were mixed with other cans and other Brands, some of them with similar colour tones, for example “Dr. Pepper´s” (reddish-violet), and some with totally different colours, e.g. “Fanta” and “Sprite”.
Different from the dummy cans, which remained unaltered all the time and only changed their places to create random conditions, the colour of the CocaCola cans had been altered in steps of ΔE. They were mixed with some more altered and some less or unaltered CocaCola cans. The expectation was that the recognition of the more altered cans will be lower than that of the less or unaltered ones.
In the end, all the cans have been presented in two different set ups: in a shelf with random arrangements of the CocaCola and other (dummy) cans, and, secondly, on a band-conveyor with the dummies and CocaCola in a random running row.
In the shelf, the participants were given some seconds time, to count the CocaCola cans, presented. On the conveyor they had the same task, looking to a certain section, while the band was running. The number of CocaCola cans, correctly identified, was considered as expression of the recognition.
The experiment started with ΔE deviations within the widely recognized range (0 to 6, as explained above). Independently from the colour deviation used, the same average quantity of cans was identified by the test persons (about 90%, which, on the other hand, means that ca. 10% of the correct cans were not found on average, not depending on colour tones). The correlation between the colour deviation and the identified number of cans was extremely low, and thus not relevant.
Therefore, the colour deviation (by altering the cans) was improved far over the valid limits. Until ΔE = 9, only a slight decrease of recognition was observed (from 90% to 88%), then, until ΔE = 13, a further decrease to about 83% of recognition was seen, which then remained stable until ΔE = 21, where the test was stopped. A relevant (mathematic) correlation or causality could not be found at all.
Another test was carried out using Nivea Body Lotion (well known dark blue plastic bottle) under similar conditions. The result was even worse, which may be explained with the fact that blue is more difficult to recognize than red.
Until ΔE = 14 the recognition remained stable at about 92% (with no correlation or causality between ΔE and number of bottles found), then decreased slightly to 88% (ΔE = 18) and then, again, increased (!) to about 91% at ΔE = 21 (no correlation, no causality, no plausible explanation).
Even in case of some errors in the test design (such as light conditions or timing), the missing influence of colour deviation is that big, that the findings remain still relevant.
What does this mean to us?
For the original purpose of the tests (recognition value of security features) it means, that people are widely unable to identify colour-driven optical security features like colour changing inks properly, even when asked to do so. They may see the switching effect, but will be unable to decide if the colour flop is the original one or another. This is really good news to counterfeiters, indeed!
But also printing works and marketing departments may relax, because even heavy colour deviations don´t have relevant influence on the recognition of brand products. As the test design was quite close to real conditions at the point of sale, all-clear can be given to Printers and Brand Owners. It is, of course, another story, when brand products with big colour deviations occur side by side in one shelf, which looks, most probably, ugly. But good recognition of products is observed even for ΔE values over 20!
by Prof. Dr. Hans Demanowski
Wer diesen Text im Original in der einschlägigen Fachpresse lesen möchte, dem Sie dieser Link/Text auf Russisch/Ukrainisch empfohlen.