Issue with Y, Cb and Cr image. Only Y (luma) image looks red, Cb looks green and Cr looks red. But converting the YCbCr image using ycbcr2rgb, give original image

Have an image and converted RGB to YCbCr color space, using rgb2ycbcr code. And when i try to seperate the Luma (Y) and color components (Cb and Cr), the generated images, doesnt look satisfying.

As per my understanding, if we take an image and convert it to RGB or YCbCr, the image should be similar in both RGB or YCbCr color space right ?

Issue image : "Original" is the image taken for the experiment. "New RGB" is obtained by converting "YCbCr Image" using ycbcr2rgb.

The Only Luma Y, Only Cb Blue difference and Only Cr Red difference doesnt make sense to me as per what i saw in internet, only luma should be a gray scale image, only Cb should be blue and only Cr should be red. But here i get strange image.

Code:

%load image

RGB = imread('example.jpg');

%get resolution of image

pic = size(RGB);

%create a zero matrix corresponding to the image

zero_matrix = zeros(pic(1:2), 'uint8');

%convert image to YCbCr color space

YCBCR = rgb2ycbcr(RGB);

%convert YCBCR to RGB again

new_rgb = ycbcr2rgb(YCBCR);

%Extract Red, Green and Blue components from the image

R=RGB(:,:,1);

G=RGB(:,:,2);

B=RGB(:,:,3);

%Generate image with only Read, Green and Blue components

only_r = cat(3, R, zero_matrix, zero_matrix);

only_g = cat(3, zero_matrix, G, zero_matrix);

only_b = cat(3, zero_matrix, zero_matrix, B);

%Extract Y and Cb, Cr components from the converted image

Y=YCBCR(:,:,1);

Cb=YCBCR(:,:,2);

Cr=YCBCR(:,:,3);

%Generate image with only Y, Cb and Cr components

only_y = cat(3, Y, zero_matrix, zero_matrix);

only_Cb = cat(3, zero_matrix, Cb, zero_matrix);

only_Cr = cat(3, zero_matrix, zero_matrix, Cr);

%Display RGB image

figure();

imshow(RGB);

impixelinfo;

title('original');

%Display Only Red image

figure();

imshow(only_r);

title('Only Red');

impixelinfo;

%Display Only Green image

figure();

imshow(only_g);

title('Only Green');

impixelinfo;

%Display Only Blue image

figure();

imshow(only_b);

title('Only Blue');

impixelinfo();

%Display YCbCr image

figure();

imshow(YCBCR);

title('YCbCr Image');

impixelinfo();

%Display Only Luma Y image

figure();

imshow(only_y);

title('Only Luma Y');

impixelinfo;

%Display Only Blue difference image

figure();

imshow(only_Cb);

title('Only Cb Blue difference');

impixelinfo;

%Display Only Red difference image

figure();

imshow(only_Cr);

title('Only Cr Red difference');

impixelinfo();

%Display new RGB

figure();

imshow(new_rgb);

title('New RGB');

impixelinfo();

Help is appreciated!

Answer 1

[Daniel]

It looks to me like the only issue is in the "Display YCbCr image" region of your script. imshow expects an RGB input matrix. It's rendering your Y component as if it were red because it expects input(:,:,1) to be the red value, and so on for the other components. And

If you want to render the Y, Cr, and Cb elements as if they were RGB, you'll need to start by converting them back into RGB image space, so imshow renders them properly. So you could render, for instance, ycrcb2rgb(only_y).

However, when you do this, you'll still run into saturation issues. Since Cr and Cb represent differences, they can represent either positive or negative values. Cr and Cb values of 0 actually denote a negative amount of red and blue, so when you want to create an image with only y, only Cr, or only Cb, the other elements have to be set to 128 rather than 0.

So when you're rendering your YCrCb components, you need to make two changes:

1. The "uninteresting" values in only_y, only_Cr, and only_Cb need to be set to 128, not 0.
2. You need to run ycrcb2rgb before passing only_whatever into imshow.

[Sunil]

@Daniel Thanks it is working fine now, after modifying the code as you told

[Sunil]

@DGM If we dont set the uninterested values in only_y, only_Cb and only_Cr by 128 as Daniel mentioned. I get the below image, which i think is not what i expect.

[DGM]

Yes, his recommendation is correct. It's late, and I had misinterpreted what he meant when he said "uninteresting". If you caught the last edits to my comment before I deleted it, it included a demonstration and explanation of exactly the effect you're describing. Once I saw your comment, I saw how you had interpreted what he said and realized how I had misread it. I apologize for the confusion.

For what it's worth, what I thought was being recommended was to offset Cb,Cr by 128 in order to center them. Like IA's comment below, there's the suggestion that Cb or Cr take on negative values. In uint8, they don't. They're already centered on 128, not zero, so there would be no need for an offset. I'm far too used to people recommending that chroma channels should be offset or scaled in ways they shouldn't be, so I'm quick to object.

https://www.mathworks.com/matlabcentral/answers/165091-how-to-convert-rgb-to-yuv#answer_823836

The advice you used (to fill the unused channels with 128 instead of 0) is correct, and should make sense when you consider now that the neutral axis is at [Y 128 128].

[Sunil]

@Daniel @Image Analyst @DGM The YCbCr image i get now is in 4:4:4. For converting it in 4:2:0.

Cb and Cr components of YCbCr in 4:4:4 should be scaled by 0.5 or 0.25 using imresize inbuilt command ?

Y=YCBCR(:,:,1);

Cb=YCBCR(:,:,2);

Cr=YCBCR(:,:,3);

Which is the corrent way ?

%4:2:0 conversion of YCbCr using inbuilt command with bilinear

Cb_auto_g = imresize(Cb, 0.5, 'bilinear', 'Antialiasing', false);

Cr_auto_g = imresize(Cr, 0.5, 'bilinear', 'Antialiasing', false);

or

Cb_auto_g = imresize(Cb, 0.25, 'bilinear', 'Antialiasing', false);

Cr_auto_g = imresize(Cr, 0.25, 'bilinear', 'Antialiasing', false);

Because i read that 4:2:0 scales the Chroma samples by 1/4th the Luma samples

[Daniel]

Honestly, how to convert from RGB to YCrCb is about the limit of my knowledge in the area. 4:2:0 is beyond my experience. You might want to post this as a new question to get a fresh set of eyes on it, or try asking one of the other people who answered.

[Sunil]

@Daniel Sure, Thanks for the input!

Answer 2

[Image Analyst]

I think you don't understand what the color spaces actually mean. imshow() treats a 3-D image as if it were a true color RGB image, which a YCbCr image is NOT. If you do display the ycbcr image as rgb, or course it will look very strange. And if you extract any of the color planes, for example, Y, then it's a gray scale image. You could display it as some color - whatever you want - but it could be misleading. And the Cb image, which is the ycbcrImage(:, :, 2) is not a green image, though you could display it as green or any other color or even apply a pseudocolor colormap. Though that would be deceptive. And like @Daniel said, Cb and Cr could have positive and negative values, so the least deceptive way to display them would be with [] in imshow, like (untested):

ycbcrImage = rgb2ycbcr(rgbImage);
[y, cb, cr] = imsplit(ycbcrImage);
subplot(2, 2, 2);
imshow(y, []);
title('Y Color Channel');
subplot(2, 2, 3);
imshow(cb, []);
title('Cb Color Channel');
subplot(2, 2, 4);
imshow(cr, []);
title('Cr Color Channel');

[Sunil]

@Image Analyst Thanks for the input, because of you i learnt how to use subplot, and display all the images in same figure

Answer 3

[DGM]

This answer shows one way to represent the YCbCr channels in pseudocolor,

https://www.mathworks.com/matlabcentral/answers/580287-how-to-display-y-cb-and-cr-components-of-an-image

Included in this thread are other examples representing the YCbCr channels in grayscale, commentary on the merits of either, and links to other threads with examples doing similar in other color spaces.