Homogenity of an image

Hi

I want to calculate the homogenity of images. This can be done directly in Matlab with

homogeneity = graycoprops(graycomatrix(img), 'Homogeneity')

Is that right?

Unfortunately I'm not allowed to use the image processing toolbox. Is there some formula for calculating the homogenity so that I can implement it myself?

Or is the sourcecode of graycomatrix and graycoprops available, so that I just can take it?

Answer 1

[Image Analyst]

Sure, I'm sure you can find someplace on the internet that has the algorithm.

If you have access to someone's computer who has the Image Processing Toolbox:

>> edit graycomatrix.m
>> edit graycoprops.m

[Image Analyst]

Sepp's "Answer" moved here:

Thank you for the answer.

In the matlab documentation the formula for the homogeneity is written.

I have another question. Why do we have to use grayomatrix, i.e.

homogeneity = graycoprops(graycomatrix(img), 'Homogeneity')

instead of just

homogeneity = graycoprops(im, 'Homogeneity')

for getting the homogenity of an image?

[Image Analyst]

What is im? The properties of the glcm are computed from the glcm, so you need to have a glcm first . They don't automatically calculate the glcm for you in graycoprops() probably because there are too many options to decide on in graycomatrix() and they leave those decisions to you rather than assuming something.

[Sepp]

im = img, i.e. the image.

I just want to differentiate between the homogeneity of images. So e.g. for wood images I want to get a high homogeneity value and for landscape a low value.

What graycoprops options should I choose in this case?

[Image Analyst]

I usually divide the image up into tiles and get the means of all the tiles. Look at the standard deviation of the means. For uniform images, it will be low and for variable/splotchy images it will be high. See the FAQ: http://matlab.wikia.com/wiki/FAQ#How_do_I_split_an_image_into_non-overlapping_blocks.3F

[Sepp]

This makes sense and has perhaps much lower runtime.

Do you mean the mean of all pixels in the tile?

So, if I have n tiles I get n mean values. How can I calculate the standard deviation of it?

By the way, what tile size do you use?

In the link you gave me, it is written that we can use mat2cell() or indexing to split an image into blocks. Which methods is faster in terms of runtime?

[Image Analyst]

Correct.

StDevOfMeans = std(tileMeans);

[Sepp]

Thank you.

Is the approach with mat2cell() or the indexing approach faster for creating the tiles?

How big are your tiles usually?

[Image Analyst]

Probably indexing is faster. How big you make the tiles is up to you but a common method in surface metrology or profilometry is to use a 5 by 5 grid. See standards ASME B46.1 and ISO 4287. I usually like to go a bit more since I like a few more than 25 samples. But I probably wouldn't go more than a 10 by 10 grid. Much more and you'll be approaching simply taking the histogram of the image and that really doesn't tell you about the spatial uniformity.

[Sepp]

Thank you very much.

Answer 2

[keerthana ganeshan]

WHAT IS HOMOGENITY IN IMAGE?

[Image Analyst]

Uniformity in gray levels.

Answer 3

[ahmed]

hi when i compute homogeneity and energy of images the result is very small 1.9^-20 can you help me to normalize these small values ???

[Image Analyst]

What formula did you use? Why not just multiply by 1e20 or something?