problem in ycbcr color space

I am facing some problems in using the ycbcr color space. For a certain purpose this model is the most suitable. But conversion from RGB to ycbcr yields me some negatives values too... Please help. Why does this happen? Are there any ways to solve this?

[Star Strider]

It will help if you posted (or attached) your code, and attached your image.

Answer 1

[Walter Roberson]

My speculation would be that you started with a uint8() image, used double() on that, and asked to use rgb2ycbcr() on the result. That could result in negative values, as the datatype of what you ask to convert (uint8 or double) determines which range of values are expected (0 to 255, or 0.0 to 1.0).

Always remember: the result of applying double() to a uint8 image is not an image! If you want your image represented in double precision then use im2double() not just double()

Answer 2

[Image Analyst]

I'm not that familiar with that particular color space. What conversion are you using? I see from https://en.wikipedia.org/wiki/YCbCr that there is a minus sign in the formulas so perhaps they can go negative. I assume you're using rgb2ycbcr(). What colors are you getting negative values for?

[Ashmil Mohammed]

Yes that is the function I use to change . But I am getting negative values in 2nd channel.

[Image Analyst]

If you answer my questions, maybe I can answer yours . OK, if you won't tell me the colors, then maybe you could upload the image. Uploading your m-file, or at least a snippet, would make me more likely to try it, and sooner, than if I had to write it myself.

Answer 3

[Thorsten]

You have some error in your formula. RGB values in the range 0, 255 or R'G'B' values in the range 0..1 map to non-negative YCbCr values. See https://en.wikipedia.org/wiki/YCbCr

[Image Analyst]

But in his comment to me he said he didn't use any formula -- he said he used the built-in rgb2ycbcr() function.

[Ashmil Mohammed]

will it be due to my values are 0-255 range? I think that is the case. Any how I started working on HSV so I am OK for now!

[Walter Roberson]

Try

RGB = cat(3, uint8([0 0 0 0 255 255 255 255]), uint8([0 0 255 255 0 0 255 255]), uint8([0 255 0 255 0 255 0 255]));
image(RGB);
y = rgb2ycbcr(RGB);
min(y(:))

I cannot test this on my system as I do not have the toolbox for it.

This arrangement tests all the combination of minimum and maximum RGB codes. If something is going to come out negative then it should show up here.

[Image Analyst]

Walter, the answer is 16. Ashmil, it's probably best you switched to HSV color space, though I was wondering what situation you had that made you original think "For a certain purpose this model is the most suitable". I think for most situations that I know of, conversion to HSV is fine. The only time I switch to LAB is if I need to compute a color difference because the Delta E formula in LAB space is much, much simpler than it is in HSV color space.

[Walter Roberson]

16 is the answer to be expected from the description of YCbCr in Wikipedia.