How to display Y, Cb and Cr components of an image?

Hi everyone, I've got an image which im trying to seperate out its Y, Cb and Cr components and display them. I seperated them out using the equation that relates YCbCr to RGB.

%% Reading the image as a tiff format
im1=imread('image3.tiff');
im1=double(im1)/256;
im1(:,:,4)=[];
xsize=size(im1,1);
ysize=size(im1,2);
%% Transforming the image from RGB to YCbCr
R=im1(:,:,1);
G=im1(:,:,2);
B=im1(:,:,3);
Y=0.288*R+0.587*G+0.114*B;
Cb=-0.1687*R-0.3313*G+0.5*B;
Cr=0.5*R-0.4187*G-0.0813*B;


%% Display the various different Y, Cb and Cr
figure;
subplot(1,3,1)
image(im1);

I'm now trying to find out how do I display them seperately. Does anyone know how to do it?

Thank you

Answer 1

[Stephan]

I used the inbuild example file peppers_RGB_tiled.tif, which has no 4rd dimension that could be cleared, so i used a try catch block to avoid errors:

%% Reading the image as a tiff format
im=imread('peppers_RGB_tiled.tif');
im1=double(im)/256;
try
    im1(:,:,4)=[];
catch
end
xsize=size(im1,1);
ysize=size(im1,2);
%% Transforming the image from RGB to YCbCr
R=im1(:,:,1);
G=im1(:,:,2);
B=im1(:,:,3);
Y=0.288*R+0.587*G+0.114*B;
Cb=-0.1687*R-0.3313*G+0.5*B;
Cr=0.5*R-0.4187*G-0.0813*B;

%% Display thoe original & the various different Y, Cb and Cr
figure;
hold on
subplot(1,4,1)
imshow(im);
subplot(1,4,2)
imshow(Y);
subplot(1,4,3)
imshow(Cb);
subplot(1,4,4)
imshow(Cr);
hold off

[Swapnil Srivastava]

Thanks!

[DGM]

There are a couple problems with this answer.

First of all, the luma constants in the transform look like they got typo'd The BT601 luma constants are [0.299 0.587 0.114]. It should also be noted that while YCbCr and YPbPr use the same transformation matrix, common integer YCbCr has offsets and scaling to consider. For visual purposes, it should suffice, but don't expect the values to be directly comparable to the results from something like rgb2ycbcr().

Second, since the above transformation is essentially YPbPr, the chroma channels are unit-scale and centered on zero. Because of that, imshow() will truncate them. Instead, offset them by 0.5 for viewing.

im = imread('peppers.png');
im1 = im2double(im); % needs IPT in older versions

% Transforming the image from RGB to YCbCr
[R G B] = imsplit(im1); % R2018b+
Y = 0.299*R + 0.587*G + 0.114*B; % 0.299
Cb = -0.1687*R - 0.3313*G + 0.5*B;
Cr = 0.5*R - 0.4187*G - 0.0813*B;

% Display the original & the various different Y, Cb and Cr
subplot(2,2,1) % i changed the tiling for web-view
imshow(im);
subplot(2,2,2)
imshow(Y);
subplot(2,2,3)
imshow(Cb+0.5); % offset
subplot(2,2,4)
imshow(Cr+0.5); % offset

Answer 2

[Image Analyst]

Try this:

rgbImage=imread('peppers.png');
% Transforming the image from RGB to YCbCr
yCbCrImage = rgb2ycbcr(rgbImage);
[yImage, cbImage, crImage] = imsplit(yCbCrImage);
% If you have r2018a or earlier, use the code below instead of imsplit().
% yImage = yCbCrImage(:, :, 1);
% crImage = yCbCrImage(:, :, 2);
% cbImage = yCbCrImage(:, :, 3);

% Display thoe original & the various different Y, Cb and Cr
figure;
subplot(2, 2, 1)
imshow(rgbImage, []);
title('RGB Image', 'FontSize', 16);
subplot(2, 2, 2)
imshow(yImage, []);
title('Y Image', 'FontSize', 16);
subplot(2, 2, 3)
imshow(cbImage, []);
title('Cb Image', 'FontSize', 16);
subplot(2, 2, 4)
imshow(crImage, []);
title('Cr Image', 'FontSize', 16);

[Swapnil Srivastava]

Hey, sorry for the late reply. I've tried your code and while it works, are you sure the Cb and Cr look correct? There is almost no colour in the Cb and Cr images, and from googling examples of YCbCr they look quite different to your example.

[Image Analyst]

Yes, of course they're correct. I used the built-in MATLAB functions. Like Stephan's and my answers both, the Y, Cb, or Cr image is a monochrome image and there will be no color in it. I'm not sure why you think there should be. You don't take a color image and decompose it into individual orthogonal components and have them still be color. Just think about it if you had gotten the RGB components. If you displayed the Red component alone, it would show up as monochrome (grayscale). Of course you could apply a colormap to have that monochrome red channel show up in all kinds of crazy psychedelic colors, or even just shades of red, if you want, but the underlying red channel image is still monochrome.

Do Stephan's images show up in color?

Give me a link to where the YCBCR images show up in color and I'll explain what they're doing so you understand what they did to deceive you. For example, maybe they just displayed the YCbCr image as a 3 channel image where they pretend Y is the red channel, Cb is the green channel, and Cb is the blue channel. This is very misleading and deceptive if they did that -- very hard to interpret.

My images show more contrast than Stephan's because I used [] in imshow() to expand the contrast so that you could see the channels better.

[Swapnil Srivastava]

This is the image that I was using as a reference. I suppose with your images (and mine when I tried your code), the Cb and Cr looked quite different to what the ones below look like. Keep in mind these images are not extracted from a matlab code but from a pdf, so maybe that could explain why they look different.

[Image Analyst]

They applied an artificial colormap to the monochrome Y, Cb, and Cr images. Without a doubt. I guarantee it. The images don't really look like that. Neither mine nor Stephan's look like that. If you want to you can apply a colormap but don't be deceived - the underlying image is still gray scale, not color. Actually I think doing that makes the images harder to visualize.

[Manthan Anil Agrawal]

Hello,

I am new to image processing not sure but is this correct ? looks like cr and cb got reversed.

[yImage, crImage, cbImage] = imsplit(yCbCrImage);

Shouldn't this be -

[yImage, cbImage, crImage] = imsplit(yCbCrImage);

[Image Analyst]

@Manthan Anil Agrawal, yes you are correct. Thanks for catching that and I'll fix it above.

Answer 3

[DGM]

While I agree with Image Analyst that grayscale representations of the chroma channels are accurate, easier to read, and should be more than sufficient, sometimes people want some sort of pseudocolor representations. This can either be done by colormapping or by representing the given chroma values on a locus of constant luma.

Consider the examples in the following related questions:

https://www.mathworks.com/matlabcentral/answers/303636-view-individual-components-of-ntsc-y-i-and-q-image

https://www.mathworks.com/matlabcentral/answers/1701950-how-can-i-display-i-of-yiq-from-a-jpg-image

https://www.mathworks.com/matlabcentral/answers/1600790-display-l-a-b-space-channels-separately

You should be able to take the concepts demonstrated in YIQ and LAB and apply them to YCbCr without much trouble.

RGB = imread('peppers.png');

YCC = rgb2ycbcr(RGB);
[Y Cb Cr] = imsplit(YCC);
z = 127*ones(size(YCC, 1), size(YCC, 2)); % chroma center is offset!
Ysub = 127*ones(size(YCC, 1), size(YCC, 2)); % assume some luma

% substitute new luma, zero opposite chroma channel
just_Cb = ycbcr2rgb(cat(3, Ysub, Cb, z)); 
just_Cr = ycbcr2rgb(cat(3, Ysub, z, Cr));

montage({RGB Y; just_Cb just_Cr}'); % display it

Like I said, I don't really think that's easy to intuit, and the fact that we're using a dummy luma value means it's a misrepresentation anyway. Take it as you will.

Maybe nobody will ever ask this, but I'll point it out anyway. While the grayscale representations of the luma and chroma channels can be taken and (if scaled properly) be concatenated and converted back into an RGB image, these pseudocolor representations generally cannot. Think about why that might be. Think about why it might seem like you could. After all, the pseudocolor versions of Cb and Cr just have some constant luma and are assembled using standard conversions. Why can't we just take one step backwards and get Cb and Cr and then convert back to RGB?

% try to reverse the process
YCC_Cb = rgb2ycbcr(just_Cb);
YCC_Cr = rgb2ycbcr(just_Cr);
Cbrec = YCC_Cb(:,:,2);
Crrec = YCC_Cr(:,:,3);
RGBrec = ycbcr2rgb(cat(3, Y, Cbrec, Crrec));

% show the error in the recovered image
err = im2double(imabsdiff(RGB,RGBrec));
imshow(err/range(err(:)));

Think about why it's saturated regions on the primary-secondary corners that are being distorted. What if we picked a different value for the substituted luma? Is there any luma value that would be safe?