The following article presents the steps taken in a journey by Jack Penfold, PhD, GBPS, RPSC, FBFS (@capejack) to quantify postage stamp colors, one of the trickiest elements of philatelic study. The article was submitted by APS Chapter The Stamp Forum as an Article of Distinction for 2025 and originally appeared in volume 9, no. 1 (October-December 2024) of The Stamp Forum Newsletter.

To learn more about The Stamp Forum, click here.

To read other Articles of Distinction, click here.

Steps Towards Quantifying the Colors & Shades of Postage Stamps

By Jack Penfold, PhD, GBPS, RPSC, FBFS (@capejack)

Introduction

How many times have you gone to look up a stamp in a published catalogue, only to find that the stamp has multiple color shades listed, and you have no way to know which one is yours? Many of us have found ourselves in situations like this over the years and wondered if there is a simple way of figuring it out.

Figure 1: Australia, SG17 1d Red, King George V “Sideface” definitive stamp first issued in 1913, which will be the main subject of this color study.

Recently, I have spent much time and done much research, trying to come up with an answer. As a result of this work, I have learned the following, which I would like to share with my fellow Forum members:

• The eye’s response to light is generally logarithmic, rather than linear.
• Shades are produced by changes in the ratio between the RGB (red, green, blue) colors that the eye perceives, rather than absolute values.
• How we measure the RGB values in colors and the ratios between them.

The Eye’s Response to Light

The human eye has a remarkable ability to adjust to different light levels, from bright sunlight to dim moonlight. This is possible because the eye responds to light in a way that is basically logarithmic. This has two consequences.

First, a large change in light intensity is needed for us to notice a difference when we're in bright light, while in low light, even a small change is noticeable. This allows us to see well in both very bright and very dark environments, helping to protect our vision from being overwhelmed by sudden changes in brightness [1]. Second, the faintest light that the eye can perceive is several orders of magnitude fainter than the brightest.

Perception of Colors

The way we see colors depends on the ratio between the three main colors of light: red, green, and blue (RGB). Our eyes have special cells that are sensitive to these colors, and they work together to create the colors we see.

It's not just about the quantities of red, green, or blue in the light, but how much of each color there is compared to the others. For example, if there's more red light than green or blue, we see the color as reddish. If all three colors are balanced, we see white or gray. Changes in this relative contribution from each color is what gives us the wide range of colors we see in the world around us [1].

More About RGB Values

RGB color values are measured by looking at how much red, green, and blue light is in a given region. Each of these colors is measured on a scale from 0 to 255, where 0 means there is none of that color, and 255 means there is the maximum amount. By combining different amounts of red, green, and blue light, we can create any color.
For example, if the RGB value is (255, 0, 0), that means the color is pure red, with no green or blue. If the value is (0, 255, 0), it's pure green, and if it's (0, 0, 255), it's pure blue. When all three are at 255, the result is white, and when all are at 0, it's black (see Figure 2 at right). This system helps computers and screens display colors accurately, and it can also be applied to evaluate stamps [1].

Figure 2: A Venn diagram of the RGB color relationship, in which the overlapping areas illustrate how the colors look when combined.

The displays for televisions and computer monitor screens produce an array of tiny elements, called “pixels” for short. These pixels, arranged in groups of three, one showing only red, one showing only green, and the third only blue, are projected onto (i.e., added to) a black screen. As previously mentioned, the intensity of each of the three colors is represented by a value ranging from 0 to 255.

The whole range of colors seen on the screen is then produced by varying combinations of these three numbers in all pixels on the screen. For a philatelic example, a Rose-Red stamp would be represented by (199, 45, 88) and Red-Brown by (95, 24, 34). These are the actual RGB numbers for those colors taken from scans of swatches from a Stanley Gibbons Color Key (see Figure 3).

Figure 3: Australia, SG17 1d Rose-Red, shown here along-side SG color swatches for Rose-Red & Red-Brown.

While there are different contributions from the primary colors, the red value is always the highest in this case, since we are considering two shades of red. The range of colors produced in this way is referred to as the RGB color space.

A color space is simply a mathematical model that defines a specific range of colors that can be displayed or printed. Each color space has a set of primary colors, which can then be used to create all other colors in that space. I have chosen to use the RGB model for my study, because it is already the basis for most of the other color models, whose mathematical figures are derived from RGB. It is these mathematical expressions for color that we will use to quantify the color shades of stamps.

The Case Study

For this article, I decided to conduct a case study of the Australia King George V (KGV) 1d red, line-engraved issue of 1913-1914 (see Figure 1 on the first page of this article). This was a short-lived issue with only four recorded shades (see Figure 4 below). I thought this limited issue would make for a straightforward, concise project to start a discussion on color shades.
There were only two printings of this issue, the first in November 1913, and issued in early December, and the second in February 1914, issued March. After that, production issues caused the Australian Post Office to switch to letterpress printing for the KGV Sideface Series, as it came to be known.

The 2012 Stanley Gibbons Australia Catalogue [2] lists two shades, Red and Pale Rose-Red, the latter from Plate 1 (see later). The 2021 Scott Volume 1A [3] lists only one: Carmine. The Australian Commonwealth Specialists’ Catalogue – King George V Second Edition of 2001 [4] lists four shades: Pale Red, Rose-Red, Bright Scarlet, and Carmine-Red. I chose to use the latter as the color-shade reference for this study, as it offered the greatest number of possible shade varieties. That also meant that I was choosing to accept the descriptive terms used for the color shades in that catalogue.

Figure 4: Australia King George V 1d line-engraved stamps from the author’s collection measured for this case study. In this figure, the black background has been left in place to highlight each complete stamp, but for the later color-shade comparisons, the images will be cropped to remove the black, similar to what was done in Figure 3.

Four plates were manufactured for this issue, each with 12 horizontal rows of 10 stamps. “Both printings were made from the Plates 1-4 used together, and these sheets (of 4 x 120 = 480 stamps) were subsequently guillotined into post office sheets of 120” [4, P3/2]. What makes things even more interesting is that Plate 1 was “rolled in” less deeply than the other plates, meaning that the impressions were shallower compared to those in the other plates, resulting in stamps from Plate 1 being in distinctly paler shades than those from Plates 2, 3, and 4.

This situation provides an ideal opportunity to test the ideas presented up to now on an issue of stamps which was issued in only four shades: the two paler shades (Pale Red, Rose-Red) from Plate 1, and the two darker shades (Carmine-Red, Bright Scarlet) from Plates 2, 3, and 4. My first thought was that the four shades were produced in two pairs, one pair from the first printing, then a different batch of ink was mixed and used for the second printing, producing the second pair.

Fading, Sulfuretting & Paper Color

Before describing the color-measuring process for the stamps in the case study and discussing the results, I would like to touch briefly on the issues of fading, sulfuretting, and paper color, all of which have the potential to affect color. These are important factors to consider for collectors who are interested in doing color-shade studies of their own.

In this case study, fortunately, none of the Australia KGV stamps showed any evidence of fading or sulfuretting. The stamps were printed on white or cream-toned paper, so paper color was also not an issue. That said, it is important to recognize that any of these three issues could be factors when assessing color, so please be sure to take them into account when doing your own studies.

Measuring the Colors

At this point, it would seem that all one would have to do to quantify the colors of stamps would be to measure the contributions from the R, G, and B to obtain a value for the color of any part of a stamp. The process is not quite as simple as that, however. Whether on a TV screen (in groups of three) or a printer, the pixels or dots, are very closely clustered together. The human eye does not see the individual dots, but rather the effect of the combination.

It is the relative contribution from each color, i.e., the ratio, between the three colors, to which the eye/brain combination responds [5]. At some point, as this ratio changes, the eye picks up a slight change, and a shade is born. Much has been written about what is called the “just noticeable difference,” or the amount of change in the ratio needed for the eye to perceive one shade as different from another. Studies show that the eye seems to be far more sensitive to these changes than measuring instruments are [6].

The question then becomes, “How does one measure these ratios?” The logarithmic nature of the human eye’s response to light allows for a simple answer to this question, once two requirements have been met.

What is a logarithm? A logarithm is simply an exponent. In this case study, it is the exponent of 10, which produces a particular number. For example, the logarithm of 100 is 2, since 100 = 102. The logarithm of 2 = 0.301, since 2 = 100.301. We can, therefore, multiply two numbers by adding their logarithms, and divide by subtracting them. Thus, 50 = 100/2 = 102 - 0.301 = 101.699.

First, the RGB measurements need to be taken in a “normalized” form, which means expressed as a number between 0 and 1, rather than 0 and 255. Taking the two examples shown in Figure 3, Rose-Red (199, 45, 88) becomes (0.776, 0.176, 0.346), Red-Brown becomes (0.373, 0.094, 0.133), and so on. This conversion to logarithms is a relatively straightforward process.

Second, the RGB measurements need to be performed in the “perceptual space” of the eye, since we are trying to replicate the system used up to now. This means the image file needs to be encoded in a semi-logarithmic way.

Once the RGB readings are in logarithmic form, taking the R value, as described above, and subtracting the corresponding B value gives us the exponent representing the ratio between the two colors, or the R-B Color Index. With three colors being measured, there are, therefore, three possible differences, or Color Indices, available: B-G, R-B, and G-R. These represent the Blue-Green, Red-Blue, and Green-Red ratios.

The Measurement Process

All stamps were scanned using a Canon MX-720 scanner/ printer operated via the IJ Scan Utility set to the ScanGear mode. Settings for any color adjustments were turned off in order to keep the colors seen as natural as possible.

All scans were done against a black background, at 1200 dpi, with the stamps set directly on the platen. All stamps were scanned both front and back. Images were saved as .tif files to preserve as much information as possible. The scanning area was cropped to as small a rectangle as possible, while still leaving a black border all around the stamp.

When trying to obtain RGB values for a stamp of a single color, there are two methods which can be used. The color of the stamp as a whole can be measured, or a number of small areas can be sampled. The first approach involves the use of software to remove the effects of both the postmark and the paper. For me, using small postmark-free areas for obtaining the values is simpler, as it did not involve the intricacies of 3D plotting software.

The goal of this case study is to obtain RGB values for a number of postmark-free areas of uniform color. There are a number of open-source programs available for processing the stamp image to separate it into the individual color channels.

Figure 5: Partial screenshot in ImageSleuth showing one of the Australia KGV 1d line-engraved stamps in the study alongside grayscale images generated by the software.

Currently, my preferred software is ImageSleuth [7], primarily because it can be downloaded and run offline. The original stamp image is loaded into ImageSleuth, which generates up to 75 grayscale images for various color spaces, including RGB and others. See Figure 5 above for an example of what the ImageSleuth display panel looks like.

The RGB images are saved for processing in the GNU Image Manipulation Program (GIMP) [8]. GIMP is an open source, cross-platform image editor and analysis program. I am currently running version 2.10.38, which was released on 05-May-2024.

I measured RGB values for stamps in each of the four shades, as described in the earlier section of this article, “How RGB Values are Measured.” The six sampling areas were chosen from the ten areas shown in Figure 6 at left. The RGB measurements were entered into an MS Excel workbook, which was then set up to calculate the three Color Indices. The plotting of graphs and statistical analysis were then performed in that same workbook (see Table 1 below).

Figure 6: All the potential sampling areas identified on the Australia, KGV 1d stamps. In the end, six sampling areas were chosen on each stamp.

Table 1: RGB and Color Index Data for the author’s Australia KGV 1d Line-Engraved Issue, 1913-14.

Plate No.	Stamp Shade	R	G	B	R-B	B-G	G-R
1	Pale Red	0.772	0.286	0.365	0.407	0.079	-0.486
	Rose-Red	0.806	0.355	0.414	0.392	0.059	-0.451
2, 3, and 4	Bright Scarlet	0.779	0.242	0.304	0.475	0.062	-0.537
	Carmine-Red	0.745	0.239	0.309	0.436	0.070	-0.506

 

Figure 7: Australia King George V 1d line-engraved stamps from the author’s collection measured for this case study, the same stamps as in Figure 4. This time, the black background has been digitally cropped off, hopefully making it visually easier for the reader to see the “just noticeable differences” in the color shades.

The Ternary Plot, aka The 3-Color Diagram

The fact that there are three Color Indices involved presents a bit of a problem when attempting to display the data graphically for analysis. After much experimentation, I decided not to use a 3D plotting package, even though there are many good freeware packages available.

Instead, I chose a modified technique long used by chemists and geologists, the Ternary Plot [9]. These plots are used for displaying the percentages of the constituents in solutions and in rocks.

The graph paper used in ternary plots consists of a grid of equilateral triangles, rather than the squares we are used to from our schooldays. This technique allows three variables to be plotted in two dimensions.

Figure 8: Ternary Plot (3-Color Diagram) for the author’s four Australia King George V 1d line-engraved stamps examined in this case study. The datapoints correspond to the plotted values for the Color Indices shown previously in Table 1 and to the stamps shown in Figures 4 and 7. See the next page for how to identify the unlabeled datapoints.

In the “traditional” ternary plot, the three variables must add up to a constant. For instance, the percentage of the three components in a solution sum to 100%, and all three axes are scaled from 0 to 100. Each set of data needs to be scaled so that the sum of the three is always 1, or 100. Doing this means that the three variables are no longer independent of each other. In mathematical language, it means there are only two degrees of freedom rather than three. This is what makes it possible to plot the three variables in two dimensions.

I think the situation is probably best explained using the RGB data from Table 1. Let’s say we wanted to plot the RGB data for the four stamps on a Ternary Plot. For the Pale Red, R+G+B = 0.772+0.286+0.365 = 1.423. Then, R = 54.0%, G = 20%, B = 26% of the total, and so on for the other stamps. Notice how, once two values have been calculated, the third can readily be determined, since the three must add up to 100%. Others have used this technique in philatelic studies as recently as 2010-11. Those studies inspired me to use the Ternary Plot in this case study [10, 11].

For the Color Indices, note that (R-B) + (B-G) = R-G, or –(G-R). For the Pale Red, this becomes 0.407 + 0.079 = 0.486 = -(-0.486). The three Color Indices do sum to a constant, 0. This means they are not totally independent of each other and can be plotted on a 3-Color Diagram once a suitable scale can be determined for each of the axes.

I found a tutorial on how to create a Ternary Plot in MS Excel [12] and adapted that for use with the Color Indices. Figure 8 on the previous page shows a plot of the four stamps shown in Figures 4 and 6, with datapoints listed in Table 1. Note the three axes, labelled in order, and how values on each axis always increase as one moves in a clockwise direction. With the diagram set up this way, points near the apex of the triangle have more blue in them, those near the lower left have more green, and those towards the lower right have more red, as indicated on the graph.

Probably the hardest part of using the 3-Color Diagrams is getting used to axes which are not perpendicular to each other. With that in mind, note the tick marks on the axes. These indicate the direction in which the coordinates are read or plotted.

Take the rightmost data point, Bright Scarlet. Starting from that point and moving horizontally to the (B-G) axis, the value is about 0.062. Heading upward and to the right from the datapoint, the (R-B) value is read as 0.475, and finally, diagonally downward to the right, -0.537 is read off the (G-R) axis. A check with Table 1 confirms these values. Try this process on the other three datapoints to see if you can identify the shade in each case. See [13] in the References for the answers.

Discussion

The four stamps, each of a different shade, are well separated in the plot shown in Figure 8. As more stamps are measured and added to the diagram, they start clustering around the four shades already shown. The clusters grow and eventually may overlap. There are two issues to be dealt with:

1. How to define the size and boundary of each cluster (shade).
2. What to do with stamps in a region where two clusters overlap.

At this point, the purpose is to investigate how the first issue can be resolved. To that end, a center needs to be established for each shade cluster followed by a size. At this point, it is not known where the center of each of the four shades is situated. It is probably reasonable to assume that the Color Indices of the four stamps measured are typical of the relevant shade cluster.

Therefore, I start by assuming their coordinates to be those of the center for their cluster. There is then a boundary around each of these centers, within which another stamp may be considered to be of the same shade. The extent of the boundary is defined by the Just Noticeable Difference (JND) mentioned earlier.

For simplicity, think of the four shade clusters as circles with a radius equal to the JND, with its center at one of the four points shown in Figure 8. In reality, the area is not a true circle, but let’s keep it simple for now. Next, the distances between the “center” of each cluster and each stamp measured will need to be calculated to determine its membership in a cluster.

I then plotted the data for 14 other stamps, which were also included in the study. I drew preliminary red circles around each of the four shade centers to include all of the stamps obviously belonging to that shade. I then sorted the stamps into their four shade clusters and calculated mean Color Indices for each cluster. I replotted the 3-Color Diagram (see Figure 9 below) and repeated the red-circle exercise around the new means.

Figure 9: Partial screenshot of the Ternary Plot (3-Color Diagram) for the author’s four Australia King George V 1d line-engraved stamps examined in this case study, showing circles around the color clusters defined for each shade.

The Pale Red was the only shade with enough data to establish a meaningful circle, but at least this looked like I was going somewhere. The Pale Red cluster does include an outlier on its left edge. I decided this should be ignored for now, so I plotted a new, smaller black circle around the Pale Red center. This looked good, so I drew black circles of the same radius around the other three centers.

The black circles show how the diagram would likely look once enough data have been accumulated. Note that there are some areas of overlap. I followed up with some calculations, and found that the radius of the black circles is about 70% of the currently accepted value for the Just Noticeable Difference described above.

My original hypothesis was that one pair of shades came from the first printing and the other from the second printing. The December 1913 issue was printed on paper with horizontal mesh, and the March 1914 issue on paper with vertical mesh. This difference allowed the stamps to be identified by printing, which I did by holding them up to a light source.

The flashlight app on my smartphone proved to be very useful here. Both the Pale Red and Bright Scarlet stamps appear in both printings, which suggests that my original hypothesis was incorrect. Subsequent measurement of more stamps shows that the Carmine-Red also appears in both printings.

The total of 18 stamps measured in the study, included three joined pairs. This gave me a chance to check the consistency of my technique. After all, stamps from the same sheet should produce the same results. Results of measurements for the pairs, and their images, are presented below (Table 2 and Figure 10).

Table 2: RGB and Color Index Data for the three pairs included in the study.

ID Number	R	G	B	R-B	B-G	G-R
PN01-L	0.764	0.238	0.294	0.470	0.056	-0.526
PN01-R	0.769	0.245	0.302	0.467	0.057	-0.524
GM09-L	0.807	0.326	0.398	0.410	0.072	-0.482
GM09-R	0.794	0.331	0.403	0.391	0.072	-0.463
GM12-L	0.790	0.314	0.379	0.411	0.065	-0.476
GM12-R	0.791	0.305	0.370	0.421	0.065	-0.486

 

Figure 10: Three pairs of Australia King George V 1d line-engraved stamps also measured for this case study, in an effort to show consistency of the technique by using two stamps from the same sheet and comparing the results.

As it happens, in previous years, there has been some discussion about this fact, when three different readings were produced for three stamps [14]. The three stamps had been a strip of three until just before the measurements were taken. A statistical analysis of the above results shows that the results for all three pairs are well within the margin of error. While results for three pairs of stamps does not prove consistency, it is certainly an encouraging sign that a simple approach such as that advocated in this study is viable.

Conclusions

I have presented a new method, using Color Indices to measure the ratios between the R, G, and B colors of postage stamps, to quantify their shades. It shows promise as a machine-based system that might match the well-established unaided eye system, which has evolved since 1840.

The technique was successfully used in a case study of the Australia King George V line-engraved penny red issue of 1913-1914. Analysis of three pairs has shown that the technique produces consistent results. It has also been shown that the system can reliably assign stamps to shade clusters using distances between Color Indices in 3-dimensional space.

The 3-Color Diagram (Ternary Plot) has also been shown to be useful in distinguishing between stamps in different shade groups. The analysis also suggested that the two types of ink, which produced the four listed shades, were seemingly used for both of the two printings, rather than one type for each printing.

There are still many questions to answer, including whether distances between Color Indices can be used in the same way as in other color systems. The system did show promise that using a combination of other color measures and comparison of Color Indices could help solve the problem of what to do with stamps which may fall in a “neutral” zone between two shades.

How do Color Indices differ from one scanner to another? Can a reference system of shades be developed? The system has been shown to be viable for an issue with a limited number of shades, will it be robust enough to deal with more complicated issues? Another topic of interest is how do similarly named shades from different countries compare to each other?

Finally, I welcome comments and suggestions from interested readers and fellow TSF members. I am @capejack on TSF, and I look forward to hearing your views…. Happy stamping, all!

Acknowledgements

Many people have been involved in helping and encouraging the completion of this first paper. Thanks to all of them, presented here in no particular order. There are the members of the Fanny Bay Filatelic Society on Vancouver Island. All of them have seen and provided constructive comments on various aspects of this work as it progressed. Thanks to Clive Levinson (@clivel), the author of the ImageSleuth software package, for many helpful discussions about it, and for tweaking it in response to my requests.

Peter Newroth (President, Gerald E. Wellburn Philatelic Foundation, Victoria, BC, Canada) has provided information on the printing process and conditions in printing rooms during the line-engraved era. He and I have also had many discussions on the subject of color and shades on stamps, and the need for more research on the topic. Peter, and one of my fellow FBFS members, Gordon May, were kind enough to part with items from their collections for a while to allow me to use said stamps in helping to add data to the case study.

Jan Hofmeyr is another philatelist who has encouraged me in this endeavor and has shared thoughts and ideas on the subject. Thanks also to Chris Dorn (@berylliumguy) and Stan Brown (@stainlessb) for sharing the results of studies on peroxide soaking of stamps, and their encouragement as I proceeded with the research for this paper.

References

[1] ChatGPT website, “Human Eye Light Response.” Accessed on 20-Sep-2024. Output revised and modified by Chris Dorn (TSF Newsletter Editor) and the author, Jack Penfold.
URL: https://chatgpt.com/c/66edf3ad-e74c-800c-843e-ab43c04139ea

[2] Stanley Gibbons Publications. Commonwealth Stamp Catalogue, Australia, 11th Edition, 2012.

[3] Amos Media, Inc. Scott 2021 Standard Postage Stamp Catalogue, Vol. 1A, Sidney, OH, USA, 2020.

[4] Brusden-White. The Australian Commonwealth Specialists’ Catalogue—King George V, Section 3, Second Edition, 2001.

[5] “Human Vision: Light, Color, Eyes, Etc.,” Introduction to Computer Vision. URL: https://people.cs.umass.edu/~elm/Teaching/ppt/691a/CV%20UNIT%20Light/691A_UNIT_Light_1.ppt.pdf. University of Massachusetts Amherst, Computer Vision Lab, accessed on 11-Apr-2024.

[6] McClung, Michael C. The Journal, 45, 3, 185, 1993.

[7] Levinson, Clive. ImageSleuth software, available at: https://www.thestampweb.com/imagesleuth

[8] GNU Image Manipulation Program (GIMP). Available at https://www.gimp.org/

[9] Ternary Plots. https://en.wikipedia.org/wiki/Ternary_plot, accessed on 05-Aug-2024.

[10] Kegg, Reverend Dr. Gordon. “A Simple Approach to Measuring Shades (1)- King Edward VII 1d Shades,” The GB Journal, 48, 5, 105, Sep-Oct 2010.

[11] Kegg, Reverend Dr. Gordon. “A Simple Approach to Measuring Shades (1)- King Edward VII 1d Shades,” The GB Journal, 49, 3, 49, May-Jun 2011.

[12] Scheibe, Christian. How to Plot a Ternary Diagram in Excel, https://chemostratigraphy.com/how-to-plot-a-ternary-diagram-in-excel/?unapproved=376&moderation-hash=a131be884832befd55348c4925081d03#comment-376, 2021.

[13] Datapoint IDs from Figure 8, from left to right: Rose-Red, Pale Red, and Carmine-Red.

[14] Harman, Christopher. “King George V Shades.” The GB Journal, 39,1 (12), Jan-Feb 2001.