How can I calculate euclidian distance in RGB spaces for a skin lesion ?

To be clear I want to find The Euclidean distance D which is calculated between each pixel color and the six colors ( white , red, light brown , dark brown , blue gray and black ). I converted the RGB color space to the CIE Labcolor space but I don't know how to continue . I attach my code so far and an image.

[William Rose]

@Ioanna St,

WhenI try to runthe script I get the error

Error using imread (line 349)
File "color.png" does not exist.
Error in xromadik (line 32)
rgbImage= imread('color.png'); 

Answer 1

[Image Analyst]

"I want to find the percentage of colour appearance for each one of the six colours (white , red, light brown , dark brown , blue gray and black ).What i have to do?" <== See attached Discriminant Analysis demo. It does exactly that.

[William Rose]

@Image Analyst, amazing.

[JovanS]

Ok your code was really helpful but i have some questions in order to understand how it works.

If i have understand how it works i have to enter 6 as the number of classes (each class is for a specific color from the colors above and then when i have to draw the sample pixel colors i have to draw the whole skin lesion in order to find the percentage of colour appearance for each one of the six colours. Right?

@Image Analyst

Answer 2

[Image Analyst]

Use this function: deltaE

You will need to calibrate your images with known color standards, otherwise the delta E you get is totally dependent on the lighting you use and your exposure time. See attached tutorial.

Answer 3

[William Rose]

@Ioanna St,

Since there is a file missing which I need to run your code, I will just give a few suggesitons.

1. Post the simplest possible example that demonstrates the problem or what you are trying to do.
2. If you want to compute distance between two specific colors in CIE space, use dist=imcolordiff(c1,c2), where c1, c2 are RGB colors.
3. To get a monochrome image whose gray scale values represent color distance (in CIE space) between corresponding pixels of 2 images, use im3=imcolordiff(im1,im2), where im1, im2 are RGB images.

See imcolordiff help for additional options and examples. Good luck.

[Image Analyst]

Actually, looking at the documentation it looks like deltaE is the 1976 version, which uses Euclidean distance, whereas imcolordiff uses the CIE 1994 or 2000 definition of delta E.

imcolordiff is more accurate than deltaE when compared to human vision, and is probably what she should use. However, the numbers are really relative and arbitrary if there is no calibration standard used. They assume you're using sRGB color space (very unlikely) and they just use "book formulas" assuming sRGB. So you'll get numbers out but they won't correspond to what you'd get from a real spectrophotometer instrument.

[William Rose]

@Ioanna St,

@Image Analyst is a real expert, so I recommend their advice. 🙂

[JovanS]

Thank you for your response , but could be please more specific in the steps that i have to follow in order to find the presence of the six different colors , that i mentioned before , in this skin lesion ?

I am sorry but I am really very confused. @Image Analyst

[Image Analyst]

@Ioanna St see the tutorial I attached in my answer. Also see

colorChecker

https://calibrite.com/us/product/colorchecker-classic-mini/?noredirect=en-US

[JovanS]

I do not have a system or camera to calibrate, I just have two already taken images and i want to find the percentage of colour appearance for each one of the six colours (white , red, light brown , dark brown , blue gray and black ).What i have to do? @Image Analyst

[Image Analyst]

You can train it and classify it with a discriminant classifier. See attached demo and adapt as needed.

Or you could use k-nearest neighbors.