What is the background theory of this Skin Detection in RGB code?

Hi, I am new to MATLAB and here too. I would like to know more about how the following code works and its background theory.

Currently I am doing my Bachelor Thesis about gesture recognition. For skin detection I found the following code on YouTube whose author is Anirudh Sharma ( https://www.youtube.com/watch?v=gvtqSJx3Y98 ). The most important part of the code looks like this (based on my implement):

vid = videoinput('winvideo',1,'RGB24_640x480');
... %Some webcam setup code after this
data = getsnapshot(vid);        
I = imsubtract(data(:,:,1), rgb2gray(data));
I = medfilt2(I,[3 3]);
I = imadjust(I);
level = graythresh(I);
Ibw = im2bw(I,level);

I know it is unusual to use RGB color space to detect skin instead of YCrCb, HSV, etc... but turned out it worked pretty well, fast and robust - especially with a monotone background. But the point is, how could the skin-toned region appears clearly on the background after subtracting the Grayscale from the Red plane (I = imsubtract(data(:,:,1), rgb2gray(data));)? Is there a theory background of this?

So far I haven't found yet any reliable document about this. I also have no hope contacting the author as his video was uploaded 2 years ago and his email or Google+ on his YouTube account are unavailable. I know some people have already asked about this code before but none of them satisfy me. Therefore it would be great if there is an theoretical and practical explanation about this.

Thank you so much in advance.

Answer 1

[Image Analyst]

It can work in RGB color space for certain controlled situations, like skin tones that are similar and illumination levels and color that are consistent. It's not the most robust way to do it but it can work provided everything works out just right.

But I'm not sure what your question means. "how could the skin-toned region appears clearly on the background after subtracting the Grayscale from the Red plane" What does that mean - appears clearly? Do you want to mask out the non-skin so that it's black? Or what???

[Thanh Nguyen Bui]

First of all thank you for replying. I should have included some images to make it clearer. Let's me explain again:

First of all I have a RGB picture:

After I = imsubtract(data(:,:,1), rgb2gray(data)); I got this:

After applying Median filter and imadjust:

That is what I meant "how could the skin-toned region appears clearly on the background after subtracting the Grayscale from the Red plane". Why most of the background turns black and the skin region appears the clearest? Then why do we have to use the grayscale and the red channel, subtract each others to that? And why not use the blue or green channel?

Hope it is enough for you understand my question. Looking forward to your reply.

[Image Analyst]

If the hand is pink or reddish, then it will have a high signal in the red channel. However the white background will also have a high signal in the red channel which makes the red channel the lowest contrast color channel you could use. Look at the green or blue channel - they will have more contrast than the red channel. So then if you want to mask it, just threshold and mask

mask = greenChannel < someThreshold; % Find dark pixels.
greenChannel(mask) = 0; % Set dark pixels to pure black.

[Thanh Nguyen Bui]

Sorry for being replying late, I was working hard on my thesis today. Your answer makes things make sense now, but honestly I don't fully understand the whole "However the white background will also have a high signal in the red channel which makes the red channel the lowest contrast color channel you could use. Look at the green or blue channel - they will have more contrast than the red channel". Could you give any graphical explanation?

By the ways, I played around with the code and came up with this. Is that related to your explanation?

Red channel image:

Grayscale image:

After imsubtract:

[Image Analyst]

Well look at your hand. See how it's the same color (intensity) as the background? Try it also with the blue channel and you'll see the hand won't be the same color - it will be darker. Also post your background image (the one without the hand in there at all).

[Thanh Nguyen Bui]

Sorry for the long delay, it took some time for me to came back to the lab room with the same background. Yeah after some more experiments I got your point. Here are what I got:

Red, Green, Blue channel of the background:

It is true that they have almost the same intensity, while the hand is not. Now with the hand included:

Subtracting the Grayscale from the Green channel

Subtracting the Grayscale from the Blue channel

The hands are not visible at all. So can I say that it isn't actually "Skin Detection" but a kind of Background Subtraction that leave the hand visible?

Anyways, thank you so much for your support. I now have clearer view of what I am doing.