Can we use standard deviation of color image as a contrast value?

Hello,

Is it possible to compare the contrasts of 2 image by comparing std2(rgbimage) or is it necessary to convert to gray scale? Can you describe what is the difference between them?

Thank you

Answer 1

[Walter Roberson]

That particular kind of contrast is called RMS contrast; there are other contrast measures as well.

https://en.wikipedia.org/wiki/Contrast_(vision)#RMS_contrast

YourImage = imread('flamingos.jpg');

c3 = std2(im2double(YourImage), [], [1 2]);

would give you the contrast of each color pane independently, giving you a 1 x 1 x 3 result.

But then what do you with the vector of 3 results? You could interpret them as an RGB color

image(c3)

but does it make sense for "contrast" to be interpreted as a color??

If you examine the other kinds of contrast measures, they mostly work with luminance -- requiring that you convert to grayscale or Lab or similiar first.

[dila suay]

Thank you Walter. When I convert to Lab space, How can I find the contrast?

[Walter Roberson]

L_A_B_ = rgb2lab(YourImage);
L_ = L_A_B_(:,:,1);
%L_ is now perceptual luminance
L_Contrast_one_definition = max(L_(:)) - min(L_(:));
L_Contrast_another_definition = std2(L_);

but.... you really have to be careful about the units.

The problem with IM_contrast = max(IM(:))-min(IM(:)) is that even I have 1 pixel of black and white, it will give me 'full contrast'

Yes, and it is correct to do so. At least one definition of contast involves the difference between the brightest pixel and the darkest pixel.

You have to understand that contast is defined over the entire image -- it is a non-local calculation. Suppose you have a grayscale ramp, 0 on one size, 255 on the other, a straight linear increase, then the difference between any two adjacent pixels might only be 1, but the contrast is brightest to darkest, 255 to 0, same as if you were to randomize the pixel order or you were to have 0, 255, 1, 254, 2, 253, and so on. contrast in this sense is not measuring abruptness of edges for example.

It would not surprise me if there is a measure of abruptness of edges, of local changes in luminance. However, I do not know what that would be called.

[dila suay]

Thank you so much

[Walter Roberson]

Thought experiment:

Are the contrasts of all of these the same, or are some different?

[255,   0,   0, 0, 0]
[255,   0, 255, 0, 0]
[255, 255,   0, 0, 0]
[0,   255, 255, 0, 0]
[0,     0, 255, 0, 0]

Answer 2

[Sulaymon Eshkabilov]

Numerically to find the differences of the two imaes (of the same size) and then compute the std of the computed differences are feasible, e.g.:

A = imread('A_RGB.png');  % Size is 1000-by-1200 RGB image
B = imread('B_RGB.png');  % Size is 1000-by-1200 RGB image

diff_AB = A-B;
std_diff=std2(diff_AB);   % STD of the difference    

[dila suay]

i have more than 2 images. But i want to learn if std2(grbimage) is same with std2(grayimg) and can it be used as a contrast value?

[Sulaymon Eshkabilov]

if you're looking for an RMS contrast value, then you can easily implement the calculation equation given here into your code:

https://en.wikipedia.org/wiki/Contrast_(vision)#RMS_contrast

Which is applicable for RGB and grey images.

Simple calc of the grey image contrast value is:

IM = imread(GImage);

IM_contrast = max(IM(:))-min(IM(:));

[dila suay]

The problem with IM_contrast = max(IM(:))-min(IM(:)) is that even I have 1 pixel of black and white, it will give me 'full contrast'. Do you know how can I set a threshold for pixel count?