Accessible Color Identification Discussion

A few days ago, Don/51studebaker brought up some concerns about the topic of color identification:
https://goscf.com/t/44769&whichpage=2#380164

I would like to explain what I have learned concerning the development of a color identification method utilizing consumer scanners.

Computers are perfect at one thing, consistency. The output of a program will always be the same if the input is always the same. The major problem in using them for color identification is is that achieving consistent input has many variables. The next problem to be solved is to write a program to quantitatively analyze the input and then output a color name that is meaningfully accurate. However, accurately comparing stamp colors via scans can be done with only minimal expense in the absence of a special program.

So far in my testing, 16 bit per color channel consumer grade scanners are sensitive enough to provide useable input. However, no two scanners are exactly the same. There will be at least minor variation even between two pulled off an assembly line at the same time. Then there are differences between different makes and models in how they physically acquire the scan.

The most significant differences can be reduced to an acceptable amount by utilizing a color calibration target (~$20). Maintaining a consistent work flow that correctly utilizes color calibration is tedious when done manually. However, it could be made to be quite painless by having the entire workflow in a single program. Incorporating the open source program LPROF (http://lprof.sourceforge.net/) with correct presets would require of the user to only click on 4 landmark points of a color calibration target image.

Some of the variation in scans will still remain due to things such as different light sources. The most significant one is due to color metamerism, i.e. colors not appearing the same under different light sources: https://en.wikipedia.org/wiki/Metam..._%28color%29 . I haven't done enough testing to determine if this is a significant issue or not. If it is, then metamerism can be eliminated by only using scanners which use a specific light source. It seems that the usage of LEDs is superior and is being adopted as standard. Image sensor type is another variable. You can read more about some of the issues if you like here: http://www.imageaccess.de/?page=Whi...persCCDOrCIS . Even if only one subset of scanner type is suitable, many makes and models are available for purchase at affordable prices (below $300). As time goes on, the quality of scanners will increase and their prices will likely decrease.

Following the above, we now have acceptable input data. More specifically, by color calibrating, we now have a scan that is in the Absolute Color Space. The concept of the ACS is why any further variation introduced by a video card, monitor, etc. does not matter. Any image that is in the ACS will display exactly the same as any other image in the ACS on every single monitor on the planet (see note 1 below for an uncommon caveat to this). Even without a specialized program, their similarities or differences will be accurate if both are viewed in the same program (Firefox, Internet Explorer, Photoshop, etc.). If the monitor you are using displays all images as slightly bluish, then identical colors in the ACS will appear as identical on that monitor. Similarly, all non-identical colors in the ACS will appear as non-identical on that same monitor.

At this point, no special program is needed. A color calibrated image is all that is needed to accurately compare one stamp image to another color calibrated stamp image using your eyes. It is here that different people's ability to perceive color differences and the quality of the monitor can be an issue. However, none of that matters if you use an image manipulation program's (Photoshop, paint.net, GIMP, etc.) color picker tool to compare the numerical RGB or HSV values of each stamp with each other.

With a color calibrated image, an individual's color perception doesn't matter to a program that goes over each pixel in an image and does meaningfully accurate analysis. This will require the program to do some statistical analysis, compare the results to numerical thresholds, then output a color name. These thresholds will need to be derived from analysis of many color calibrated scans of stamps that have already been identified as a particular shade. This is the most difficult part of computer driven color identification development.

I've been working with Jack Daley (3cent1861.com) who has had good success in shade identification of the 1861 3c series by analyzing 9600 dpi scans (close to 1 pixel per ink particle). He subjectively identifies (expert eyeballs) shades based on the the ratios of the handful of pigment particle types. These particles vary by size and also color categories such as reddish, orangish, bluish, and greenish. I'm working on integrating computational quantization to further utilize his findings. In essence, I am trying to computerize his color perception and knowledge. Unfortunately, our progress is at a snail's pace due to our limited availability.

From what I can tell, Jack Daley can't place every single stamp into a particular shade category. Stamps can be faded or stained. During production there are day to day ink formula variations, containers of ink not thoroughly mixed each time they are accessed, one side of the plate receiving more ink than the other, etc. that all lead to how every stamp has a slightly different appearance. Some stamps simply have a color that is between two shades. However, just like Mike McClung, he has found identifiable patterns correlating to the production dates.

In conclusion, 30 years ago the concept of today's smartphone was inconceivable. Today's consumer grade scanners are now accurate and affordable enough to help bring definitive color identification to philately's masses. The use of color calibrated scanners (~95% accuracy) will prevent accurate identification of a small percentage of stamps, but it would allow for orders of magnitude better accessibility than now. We have everything we need except for a sufficient amount of data and a program to simplify the process for the average user.

There are many more details than discussed here and many of the concepts might need more simplification to understand. Ask away, but it may take me a long while to reply.

Note 1.

However it is only true if every computer recognizes and applies an image's color profile correctly. A color profile defines how a computer correctly displays an image. To explain this further, look a few weeks back to this thread: https://goscf.com/t/43719 . The reason why the image appeared differently on Don's two computers is because nl1947's image did not have a set color profile. One of his computer's programs defaulted to one color space and the other computer defaulted to a different color space. This doesn't occur if an image has a color profile applied and the viewing program (Internet Browser, image viewer, Photoshop, etc.) utilizes the profile correctly. This tedious part of color management can be eliminated by having one program perform the entire chain from scan acquisition to analysis.

Further reading:

http://postalmuseum.si.edu/research...orimetry.pdf Philatelic Shade Discrimination Based on Measured Color

Some scanners utilize infrared light to detect dust. This feature could aid the shade identification process:

http://www.heritagesciencejournal.c...7445-2-8.pdf Infrared False Color for Art Examination

http://chsopensource.org/2013/02/13...up-workflow/ Identification of pigments by multispectral imaging; a flowchart method

Ryan, thanks for the great information. I'm sure there are many who can understand it all. It's tuff to teach an old dog new tricks. It would be great if a program were available which when a stamp was scanned, could determine the color. For someone like myself, it's an issue at times because I have certain color blindness. Please let me know if you have any suggestions which are simple. This old dog is willing to try ///

Thanks again for the thread.
Don

Hello Don,
What I propose won't eliminate the need for expertisers with very expensive colorimetric devices, but will fill most of what is a great gap in need of filling.

I and others striving for the same thing need more data. I.E. we need more scans of certified color stamps next to color calibration targets.

Unfortunately, a simple solution doesn't exist yet. In the mean time I will gladly dedicate my time to do the dirty work of color calibrating the scans of anyone who is willing to obtain a color calibration target (~$20) and scan their stamp next to it. This will allow us to create a database of known color rarities to compare to. I'll even dedicate my time to place the color calibrated image of ANY stamp that I help create next to an image of a known color rarity so that they can see for themselves how much their stamp compares and contrasts.

As for helping you with your issue of color blindness, I hope that ray.mac while chime in with helpful information. Despite his type of Red-Green color blindness, he is a successful student of the 1861 3c series shades. It's quite impressive really. I can give you a few pointers of how to use a color picker tool to compare two color calibrated stamps next to each other, but Ray probably has better tricks up his sleeve.

I didn't mention it in my original post so I'll mention it now: the methods I describe are applicable to ALL stamp issues and colors. The 3c 1851 and 1861 are probably the most complex when it comes to the number of shade variations. Other issues with only one or two shade variations will be much more easily identifiable with my described method.

I get a little nervous when you get up to 9600 dpi, as some scanners may have chromatic abberations at that magnification.

Of course it is a significant issue. What you are doing is looking at reflected light and different light sources will send in a different spectrum of light, so the reflected light will be different with different sources.

It should also be mentioned that a scanner can't just be color calibrated once and then forgotten - as they age (the light source and the detector), the color output drifts.

eyeonwall, that is a very valid concern. I did not consider it until you mentioned it. I'll spend time in the near future investigating the phenomenon and how it compares to microscopy. Until the concern is fully addressed, I want to point out that consumer scanners have been advertised as "capable" of 9600 dpi for at least a few years now.

Chromatic abberation at that level of magnification is very much a significant issue and I imagine that the engineers responsible for such high resolution scanners did at very least take the phenomenon into account. I also imagine that they found ways to overcome it in increasingly better manners over the last few years. I cannot access internal corporate documentation of how the various manufactures address the problem, but I surmise that at least one journal article/white paper can be found that details how the problem is addressed.

I thank you for pointing out that very valid concern. It also raises in my mind the question that USB microscopes might be better suited as input devices. Color slide sized calibration targets are available for not much more money, but it needs to be determined if they can be fully captured as an image that is suitable for color calibration. Until I do more research, it seems that USB microscopes don't have the focal width to capture common/cheap slide calibration targets. Give me some time and I will further investigate the issue.

Further reading:
https://en.wikipedia.org/wiki/Chromatic_aberration

I agree that is an issue, but how much so of an issue is not yet known. Further testing will determine how significant the issue is. If the issue is significant enough, then the list of suitable consumer scanners can be limited to a subset such as LED light source based scanners which seems to be the next logical progression in scanner technology.




Yes, this is an issue and I didn't address it in my wall of text for the reason that I was trying to focus on the fundamentals in hope of not overwhelming the readers of my post. I tried my best to condense the basics of the theory of my methodology, yet my original post is still very long. Yes, periodic calibration is necessary. As far as I understand it, repeated calibration needs to be done approximately every month for results suitably accurate for the intentions of general color identification. With an LED light source, the frequency of needed color calibrations is drastically reduced due to their color stability. Regarding color output drifts, they are eliminated if a stamp is submitted to me or an eventual color identification program if they are imaged next to a color calibration target every time.

Would not calibration also be needed for any monitors being used to display at the images? Not only are there potential color shade differences based upon ambient light sources on any given monitor, but monitor colors might change over time. (As they age, things like LED backlights become a bit dimmer.)

And would this method specify a specific monitor/scanner/video card/software configuration? Or do you anticipate that calibration process would greatly eliminate the need to sync up on model/brand types? If the desire is to support different brands/models do you anticipate supporting both LED monitors and the older CRT monitors commonly still found?

Do you anticipate supporting 64,000 colors like most computers currently support? While it may be possible for a group of us to agree that a particular stamp shade most closely matches a shade, having a method which guarantees a exact match to one of 64k color shades seems to be a more difficult task.
Don

I don''t understand why everyone is concerned whether or not they can calibrate their monitor/scanner to be in sync with everyone else's... As Don (51studebaker) has pointed out on earlier posts many, many times, it would be impossible to have ALL scanners//Monitors on the same page...You have to start thinking out of the box..There are ways to get a colour say from my stamp to anyone on the forum where as you see exactly what I see...and VS.

Robert

Here are a few of the requirements as I see them.

Standardized Color Chart – A standardized color chart would be needed which hobbyists could buy. Everyone would need to repurchase this chart periodically since the color chips can change over time (once every few years?). A new one could be developed which would support specific color nomenclature (existing color shade names) or NRE charges saved by utilizing one of the existing color matching system (i.e. Pantone). All hobbyists and catalogs should then reference only the color matching chart names to avoid confusion. If catalog manufacturers were unwilling to comply with the color matching naming chart, then some sort of cross reference would need to be developed. The color matching chart would become critical in the calibration procedure.

Calibration Procedure – A standardized calibration procedure would cover all the equipment required to gather and display the stamp's color shade. This would include any device used to gather the image, process the image, and display the image. The desired final result would be to be able to sit the actual stamp next to the monitor and have the shades exactly match. The procedure should cover any deltas in the scanner model, scanner driver, computer video card, video driver, monitor, and any monitor drivers. The procedure would be run once a month or so (actually also depending on total 'in-use' hours on the equipment) and be performed on each system.

Perhaps in the beginning the first generation system might standardize on the hardware. (While it is theoretically possible to have it cover all hardware models, the first generation of the system would be easier to achieve with identical equipment.) In other words, a group of folks who have the means to purchase specific scanners, computers and monitors would come together with identical systems as the proof of concept system. Once all issues have been resolved, the second generation system may evolve to include disparate hardware platforms.

Some efficiencies may be gleaned by taking advantage of existing technologies such as ICM (Image Color Matching). Licensing and other IP consideration would be required if headed down this path. I think this is a 'doable' effort, but would be at a significant cost and effort.

Perhaps a group of folks, once this is in place, could offer color matching as a service?
Don

Don..
There is a simpler way for say you to have a colour in front of you that matches my stamp exactly, and not relying on calibration of monitors/scanners (not using even a computer to guide you.)of everyone on this forum...

Robert

Robert,
Yes, this could be done non-computer aided. Let's say you and I want to color match some early Canadian stamps. We each have a large lot of the same stamp issue and desire to perform a color shade analysis.

We could easily each buy the identical Pantone color matching charts and then use our eye balls to match the colors to it. We might have a bit of 'give and take' discussions but this is because no two humans have the same eye balls. But in general, and after some negotiations, we should be able to settle upon the names of the color shades for each of our stamps. Of course this implies that we also exchange our stamps with each other via snail mail. We would each also probably have to have similar ambient lighting conditions. (Agree upon the light source in the room for each of us.)

Between two people this feasible, but it gets logistically very difficult if more people are involved. Adding back in computer technologies, where we deal with stamp images, makes supporting a larger group people easier (buts adds new overhead).
Don

Hello Don,
Needing only the scanner to be calibrated is due to a conceptual issue, not a technical one. For example, take an image of any stamp and separate it into two images by splitting down the middle. These two images now accurately represent two stamps of identical color. They will appear as the same color as each other no matter what hardware that a computer uses to display them. It's the same reason why the left and right side of any stamp image appear as the same color on any monitor. Graphics card settings, monitor settings, monitor technology (LCD, CRT), lack of color calibration, etc. all do shift the color that your eyes see at the end result on your monitor. That shift is always consistent (linear or not) for every color and hardware used to display a digital image. Likewise, two non-identical colored stamp images will always appear as having different colors.

Comparing two color calibrated images on a monitor is still imperfect due to an individual's color perception ability, really bad monitor or graphic card settings, etc. This is where a color identification program eliminates all variation except for minor calibration differences of the input sources. The numerical values for the Red, Green, and Blue channel of an image's pixel will always be the same, thus allowing for numerical comparison to another pixel. Obviously stamp images are more complex and consist of many pixels of slightly varying RGB values. The analysis of the range/ratios/etc. of colors of one stamp versus set thresholds is the most difficult part concerning developing a program. This link from my original post goes into detail about a working method for just such an analysis: http://postalmuseum.si.edu/research...orimetry.pdf

That article also demonstrates that achieving a direct match is impossible. The natural variability in stamp shades can however be put into categories of similar qualities. Not every stamp will be able to accurately be placed into a category. The non-perfect input method of color calibrated consumer grade scanners combined with fading and etc. will always result in some uncertainty that needs to be accounted for. To do that will require setting category thresholds with distinct and separated boundaries.

Hi Ryan,
Ah, I understand what you are saying regarding splitting a stamp into two images.

I still struggle with color shades and the 'human factor'. First, humans can do some strange things to a stamp. A 'new' shade pops up, who knows the history of this stamp and its handling? Who can say for sure that this new shade is not a color changling without some further chemical analysis? Would we soon begin to value 'original owner' stamps (like antique autos); those which we could be sure of the one-owner history? Mint stamps straight from the post office and never really handled/dipped as being the color standard for that issue?

And then we might also need to deal with the time element. How long would a color analysis be good for? If a stamp is analyzed as a certain shade in the year 2000; will it always be considered that same shade? Or was it a certain shade in 2000 and currently is a slightly different shade due to poor storage or handling?
Don

Don, many good points and questions.




I imagine that this would be dealt with in a manner similar to new plate varieties appearing. Though it would take much more than another stamp or two popping up for me to be convinced that a new shade variety has been found. Things like multi-spectral imaging could be used by expertisisation services to weed many of them out. I also think that things like multiple month year covers and some early documentation should exist that at least alludes to such a variety existing. I imagine that many of the claims will go the way of the China Clay Paper issues.




This one might take time to answer. Over say 10 years, the amount of color change due to aging might be apparent on a subsequent calibrated scan. Also, once nearly all of the color certified stamps are sampled, then we could determine the numerical values that identify the subgroup consisting of the best preserved/most saturated when printed/etc.




Once correctly identified, I think that a stamp should always be considered the same shade. However if its color is damaged or ages much faster than its more well preserved peers, then its value should go down just like any stamp damaged due to faults from mishandling and etcera.

It appears to me that the expertisisation of stamps is shifting to more objective quantifiable identification and condition assessment. Regardless of efforts by myself and others, I believe that the time has come where color expertisation should adopt quantification.

Much more debate and discovery will be needed to help grow this aspect of philately. I don't pretend to know all of the answers, but I do believe that this aspect can be matured without wandering into absurdity.

Historical DNA Collector, and my buddy Don to..You guys are probably geek driven as I am..And both you guys have and understand the problem with colour identification as far as sharing that colour with others...Monitors are different..PERIOD...Scanners are different..PERIOD.
My idea all along and Don I believe has listened to me will remove the failings of computers..monitors and scanners..Everyone when I suggest a colour of stamp I am looking at, will see the EXACT colour I am trying to present..The exact colour.

Robert