True color detection from taken picture

Hello I have a problem with a project, I am writing here the exact task for it:

True color detection

The red color in the ring has real RGB values: 255 , 0 , 0 . In hexadecimal form: ff0000 . The goal of the assignment is to program a program that will be able to recognize the true RGB values of the colors of the swatch. All RGB color values are shifted because the swatch was taken with a phone which distorts. I need to find out the color and then shift it by the necessary RGB values. (note: it is not allowed to use graphic programs, e.g.: Photoshop, gimp, etc. for the final realization).

I have tried it with a code which I will add here, but it only works when I load the original picture too. I need to rework without the need of the original picture,it must detect the real color from a taken picture and shift back the color value to the original picture. I hope it is understandable and someone can help me with this, I need this ASAP, thank you.

PS: You need to open both .m files (project and readPoints), but run only the project file, it opens the photo.jpg and a coursor appears, you need to click on a random color, then a point appears where you clicked, the results will appear in the the command line, everything else is explained in the scripts.

[Image Analyst]

Thanks @DGM

Answer 1

[Image Analyst]

This is a lot of work. Basically there is no "true" RGB values for the chips. The best you can do is to assume the "true" colors for the chips is the sRGB values given in the mnaufacturer's specification sheet. So then you can measure the actual RGB values of all the chips, and compare them to the sRGB values and develop a transform to change the bad/test values into estimated sRGB values. I go over this in my attached seminar.

See my Image Processing Tutorial in my File Exchange if you want code to get the sRGB values and create a chart with the sRGB values:

Image Analyst's File Exchange

I also have attached a class that helps do some of the color correction stuff and color calibration stuff. Color correction (RGB-to-RGB) is different than color calibration (RGB to CIE LAB).

[Image Analyst]

There is a colorChecker function that you can try on your image. However it may not find them right. I submitted a bug report. Code is attached.

You can make up the arrays of true RGB values and measured RGB and then use polyfit to create a transform to convert measured/test into fixed/repaired.

[Image Analyst]

Well I know how to do it and I've done it many times. It's just that it does take time. And all my existing code is very much customized to my specific images and situation. I don't have any general purpose, generic demo code for it (yet). Usually I only devote 5-10 minutes per question and to create a turnkey solution like this would take much longer than that. Maybe I'll do it this year if I have time. It would make a nice demo. Good luck though.

As a bonus though I'm attaching a background correction demo that you may find interesting. All images taken with cameras need background correction. Quite often the lens shading is quite severe - much more so than you would expect.

[Image Analyst]

I need both the .m file and the .fig file to run it.

[Jozef Uhnák]

Hi, I have a similar problem, but I am doing it in my own way with variations, I am using only the transformations and I read your seminar presentation Background correction, color correction, and LAB calibration seminar (Generic).zip about color correction and wanted to ask you if you have a resources or link to websites from where did you get those transformations from RGB to RGB (the gold standard and estimated ones), I want to learn about them more but I couldn´t find it on the internet, so I am asking you if you have the exact resources or link to websites (that would be the best) from where did you get them, I would be really thankful. The second question is if you are using RGB to RGB correction in your lattest added file ColorCheckerChartCorrection.m or something else? Thank you in advance.

[Image Analyst]

I don't have a published reference. It's just such common sense that it's nothing anyone could get a paper published on. I have however validated the concept with a half dozen of the world's top experts in color science.

The demo I attached does color correction (RGB-to-RGB) only. Not color calibration (RGB to CIELAB).

The best places to learn about color science is Wikipedia and the Educational or blog sections of the main color science companies: X-rite, Datacolor, Hunterlabs, and Konica Minolta.

And various other sites.

https://www.drycreekphoto.com/index.html

https://slightlyblue.com/

https://www.fourmilab.ch/documents/specrend/

https://calibrite.com/us/product-category/photographers/?noredirect=en-US

https://www.rit.edu/~jasvppr/1993_TAGA.pdf

http://cie.co.at/publications/technical-reports

http://poynton.ca/

http://www.brucelindbloom.com/

https://www.rapidtables.com/web/color/RGB_Color.html

https://ed.ted.com/lessons/how-we-see-color-colm-kelleher#review

and a very fun one: https://labs.tineye.com/multicolr/#colors=44ab8c,e73843,331728;weights=33,34,33;

https://www.dpreview.com/learn/?/Glossary/

https://www.cambridgeincolour.com/tutorials.htm

I could go on forever...

[John D'Errico]

Very much the same when I was working on color management problems at EKRL, and on the Colorsense program. Of course there are also degree granting programs (RIT, for example.)

Answer 2

[Image Analyst]

Attached is a demo to take a "test" (bad) image of a color checker chart and correct it such that the chip values are much closer to the standard, published sRGB values that come from the manufacturer's specification sheet that comes with the chart. This color correction (not calibration) is a transform of a bad chart (upper left) using an RGB-to-RGB transform to a corrected/fixed/repaired image (lower left) that has chips that are now closer to the standard sRGB values (upper right) than the original image.

Of course I just corrected the chart image but once you have the transform you can now use it to transform ANY color image as long as it was taken under the same conditions as your chart image.

[Image Analyst]

Yes it could work with any image provided you changed the part ofthe code where it reads the chip values because some other image may have the chart in a different location and be of a different size so you'd need to customize it to read the colors from the chip locations in that particular images.

I only call a few of the methods in the calib_utils class. If you want to delete the methods that are not used, you can certainly do that, but there's no harm in leaving unused methods in there either. You could even extract out the methods and make them regular functions in the main script if you want.

[Image Analyst]

I'm attaching an updated demo where I apply the transform to an arbitrary image (the peppers image).

[Image Analyst]

The peppers.png image is shipped with MATLAB. It's in one of the folders of the Image Processing Toolbox. Some of the images are shown if you run the attached code.

You can of course adapt the demo to display whatever images you want in whatever axes you want.

You can use the attached code to try to find the chart using the built in colorchecker function, but like I said it's not all that reliable and they are going back to the drawing board on it.

If it doesn't work then you're back to manually specifying where all the chips are, like I did.

[Image Analyst]

No I will not. I'm shocked that you would even ask. The seminar is very generic and I've posted it many times before. I did not have a generic demo for it yet but I created one just for you. Mainly because I thought it would also be a good demo to have for the future because color correction is something that comes up occasionally. I will not and would not deprive anyone else, even your classmates, of this useful demo, and the other useful code I uploaded, just because you found it useful in your particular project. That would not be fair to them or me. What we can try to do though is to make sure we never answer anything from you ever again so that there will be no record of it. This particular thread though will remain since it has none of your code in it and there is nothing proprietary or secret about it.