calculation of mean , standard deviation and average for each pixel in an image M*N?

if i have an image M*N

if i want to making each pixel in the center of window with 31*31 size , and then compute the average and standard deviation and the mean based on pixel 's nighborhoods ? how this can be done?

i dont mean computing the mean , standard deviation , and average for the whole image?

 i wait any help

Answer 1

[Teja Muppirala]

If you have the Image Processing Toolbox, the functions

IMFILTER (can be used to do mean filtering on a neighborhood),

STDFILT (standard deviation on a neighborhood),

or the more general NLFILTER (arbitrary function on a neighborhood),

are there to do these types of operations quite easily.

Answer 2

[David Young]

To get the means, you can do

   means = conv2(Image, ones(31)/(31*31), 'valid');

Note that means(1,1) is the mean for the window centred on Image(16,16), and so on.

I'm not sure how the average is different from the mean.

To get the SDs, just extend the technique above in the obvious way, using the standard formula for the SD.

[mmm ssss]

why you use the above code?

can i know?

[mmm ssss]

this will found the whole mean as in mean2(image ).

[David Young]

I use that code because it does what you want. It does not compute the same thing as mean2(Image), as you will see if you try it.

[mmm ssss]

how can i implement it for each pixel , there is not any pointers to the pixel coefficients ?

if i want to compute it for image(10,25) , how this can be done ?

[Image Analyst]

He told you. There are no "pointers" just indexing. To find the means around (10,25) just index into the means array David created for you

meanValueAt1025 = means(10,25);

mean2() does the mean of the whole image whereas conv2 will give you an image with each pixel being the mean within a 31x31 window centered around the pixel. Those are different things.

[mmm ssss]

what is required is that the mean of each pixel Separately , then this value in addation to standard deviation for this pixel used to compute loca threshold at this pixel

by

Thresh(x,y)=mean(x,y)+k*standard deviation (x,y)

really i don't know if my question is obvious or not but i think that the answers is not enough to performe the above eq.

[Walter Roberson]

Create a loop to calculate the threshold if you want to, but you can simplify it as a matrix operation,

Thesh = meanArray + k * stdArray;

David's code is an example of making a single call that calculates everything at once instead of having to code loops to do it pixel by pixel. You can code it pixel by pixel if you feel more confident about that.

[David Young]

If something is not clear, it can help to think about a small example. Suppose your image is 5x5, and the window size is 3x3 (instead of 31x31). Now look at this:

Image = magic(5)

means = conv2(Image, ones(3)/(3*3), 'valid')

Look at the numbers printed out and you can see that means(1,1) is the mean of the top-left 3x3 corner of Image, means(2,1) is the mean of the block one pixel below that, means(1,2) is the mean of the block one pixel to the right, and so on. So it's computed the mean of each 3x3 window. Isn't that what you want, scaled up to 31x31?

[David Young]

It's true that I haven't told you how to compute the neighbourhood SDs, but once you understand how the neighbourhood means are computed I am sure you'll be able to see how to get the SDs - you just need to apply the same technique to the pointwise square of the image and look up the formula for the SD.

[David Young]

... or, as Teja Muppirala points out, there's also stdfilt.