Grayscale image - saturation value

How to calculate saturation value for a gray scale image? Is there a way or is it always zero? if zero, why? My problem here is, when I pass a gray scale image to the program...due to some activity that happens on this image, there is excess of saturated pixels, either black or white. How can i diagnose that image is highly saturated without having to make a comparison to the reference image? I am actually new handling these image statistics and any help will be really appreciated.

Thanks.

[Walter Roberson]

I do not understand what the reference is to C# ? This is a resource about MATLAB.

[jane thony]

I have edited the question. Thanks!

Answer 1

[Image Analyst]

A pixel is saturated if the pixel value is at the upper or lower limit because the sensor or the digitizer cannot take any more or less light. For an 8 bit uint8 unsigned integer image, the most common, it's saturated if the pixel value is 0 or 255. Often you can tell if the image is saturated by looking at the histogram. If you see a spike in a single bin at either end of the histogram that doesn't look right, then it's saturated.

[jane thony]

So, what I understand is if I am able to calculate the histogram[mean and SD] of the current image, I should probably be able to know the differences. I'm intending to collect many such images and calculate their average values(histogram) and see if the current image under consideration is within certain limits(if it deviates more than 10%). Is this approach right in order to know if the image has deviated ? I mainly see more added dark dots or white dots in the image in some portion which is why I need to analyze the image statistics.

[Walter Roberson]

A couple of sample images would help

[jane thony]

Unfortunately, I don't have the images with me now. I want to have the right direction pointed to me so that I can work on the approach before I hit the lab again.

[Image Analyst]

Then try this

histogram(yourImage);

Answer 2

[Walter Roberson]

It is common for Black and White to be coded as 0 and 255 respectively, and you cannot really call that "saturated".

If you only have a small number of different values in the image (say 8 or 16) then your should probably assume that it has been "painted" or otherwise deliberately manipulated to change the distribution of values, and in that case you should probably guess that it is not saturated.

If you have a broader range of values, then you could do something like take the median of the counts between (say) 240 and 254 and take the ratio of the count at 255 to that. A ratio near 1.0 would be common, but if the ratio is substantially higher than 1 then 255 is "over-represented" . You could make this more concrete by assuming a normal distribution and calculating probabilities, but you have to be careful because images are not necessarily normal distribution (and sum-of-normals would probably be pretty common.)