Using Directionality plugin

Dear Mr. @tinevez,

We are trying to do a quantitative analysis of the pore direction with FIJI plugin (Directionality). We actually are not familiar with and never used FIJI/ImageJ before but we followed your instruction on https://imagej.net/Directionality and got the result (image below).

MR2_top_histogram

We got a different kind of histogram from the one on the web. There are 3 slices while the example has 1. What are these slices and why is the histogram different from the example? Note that we are using fourier components, Nbins 90 and histogram from -90 to 90. The image we used attached below.

MR2_topflip

Thank you very much in advance.

Sincerely,
Fadhli

Good day Fadhli!

Please be aware of the fact that orientation analyses are by no means a simple affair. To get an impression of important but by no means of all aspects, you may study this report about professional approaches to global/regional orientation salience. Make sure you understand why one should speak of orientation salience analysis and not orientation analysis.

Concerning your sample image, I should like to make some remarks:

  1. The image is index-colored which is suboptimum.
    As far as I can see it doesn’t contain any color information, i.e. you should actually analyze the original gray-value image.
  2. A disc-shaped window is a good start but even better are window functions that show a soft slope, like those discussed here.
  3. The window isn’t applied in a way that the outside is zero.
    As a consequence and if the whole image is analyzed, you get a large DC-component in the orientation-result.
    The same holds true for the windowed part. Therefore, it is best to use DC-removal as discussed here.
  4. It appears as if the image is insufficiently bandlimited and it must even be be suspected that it is incorrectly sampled (digitized).
    Please make sure that the bandlimit of the imaging optics accords with the Nyquist-frequency (half the sampling frequency) according to the Shannon-theorem.
  5. An aspect of orientation analysis that is not mentioned in the above cited report is the influence of low-frequency structures on the result. Most often they are caused by uneven illumination, etc. They can strongly influence the result although they are in fact irrelevant for many investigations. Carefully chosen highpass-filtering may help in such cases.
    Concerning your image, such kind of influence isn’t very pronounced.

Here are three “Orientation Salience Functions” (OSF) computed from your sample image after conversion to 32bit gray-value:

  1. DC-removal only:
    Orientation%20Salience%20of%20%22MR2_topflip
  2. Tukey-windowing (80% flat):
    Orientation%20Salience%20of%20%22MR2_topflip
  3. Tukey-windowing (80% flat) plus highpass-filtering (limit @ 2% Nyquist):
    Orientation%20Salience%20of%20%22MR2_topflip

Please note that the orientation angle starts from horizontal (0deg) and increases counter-clockwise.

You may also have a look at a more recent and related thread.

Regards

Herbie

1 Like

Dear @Fadhli

The reason why you get 3 slices is the following: Most likely you used a colored RGB image as an input. Directionality can only work on grayscale images, so it treated each color channel (R, G and B) as an individual slice.

As for the validity of the approach for your image, please see @anon96376101 post. Directionality works well only when there are salient oriented structured in the image, as a plate full of uncooked spaghetti.

1 Like

Dear Herbie,

Thanks for your reply and your examples. First of all, i’m sorry if i seem to be a little confused because image processing is not really my field and i’m new to this kind of study.

I couldn’t find the report/papers that you were referring to and the link report doesn’t work. So I tried to read the corresponding threads and I suppose you used a different software for the analysis. Could you tell me how did you obtain the results and/or send me the papers?

Regards
Fadhli

Dear @tinevez,

Thank you for the explanation. How do I know if my images have the “salient oriented structured” (actually i am not familiar of what it actually is) in order to get valid results using directionality?

regards
Fadhli

Good day Jean-Yves,

as I’ve written in my post, the OP’s sample image isn’t RGB but color-indexed.

Directionality works well only when there are salient oriented structured in the image

Just to make things clear, this is not what “Orientation Salience” in my understanding stands for.

As I’ve clearly explained in my report, the problem with all of the approaches discussed in it, is that they don’t analyze the pure geometric property of orientation because the results also depend on the contrast of the image structures. That’s the reason why we shouldn’t speak of orientation analysis but of orientaion salience analysis. This term appears adequate because it widely conforms with the visual impression.

Best

Herbie

Fadhli,

the link is not broken!
It works and downloads the zipped report “2013_Orientation.zip”.

i seem to be a little confused because image processing is not really my field and i’m new to this kind of study.

This means that you have to learn how to do things in the correct way and my report may help you with orientation analysis.

I suppose you used a different software for the analysis.

It’s all described in the report.

Please try to find the downloaded report on your computer!

Regards

Herbie

Fadhli,

just a remark on Jean-Yves’ post:

Please try to understand my reply to his post!
(Actually I don’t agree with his statement.)

Regards

Herbie

Fadhili,
Just to clearify a difference in views:
The plots submitted are based on degrees in Radians (2 x Pi)
Yours are based on Degrees (0 - 360)

The plots summitted are based on the most intense pixels in image.
Your plots are based on the overall intensity…
Overall they both give the same overall readings.
Bob

Bob,

please try to understand the matter before posting misleading statements:

The plots submitted are based on degrees in Radians (2 x Pi)

Wrong!
2*Pi means 360deg and an orientation can only have an angle from the range 0deg to (180 - eps)deg.

As note in my post, my plots go from 0deg to (180 - delta)deg.

The plots summitted are based on the most intense pixels in image.

Wrong!
My analyses are based on a DC-free windowed image.
This is completely different from “most intense pixels in image”.

Regards

Herbie

It’s not misleading and you know it. Look at your plots own axis legends.
Bob

My plots go from 0.0deg to 179.7deg.
As indicated, delta angle is about 0.3deg, i.e.180 - delta = 179.7deg.

That’s what I know and see from my plots.

Regards

Herbie

Good day Fadhli (@Fadhli),

would you please be so kind and tell us how your investigations proceed?

Did you finally get access to my report?

Are my five remarks about your sample image comprehensible?

Please report about the state of your work.

Regards

Herbie