Obtaining Aspect Ratio and Convexity of Sand Particle Using ImageJ

Hello,
I am trying to obtain the results for aspect ratio and convexity of a single sand particle using imageJ. I have been following the usual steps, setting scale, type 8-bit, apply threshold and I have set the measurements to shape descriptors, ferret shape, area and centroid. However when I measure the particle my particle count is in the thousands.

Attached are the initial SEM image and the output image from ImageJ.

If any one has any knowledge of how to solve this problem and use ImageJ for this single particle It would be very much appreciated. Many thanks.

Good day,

evidently you need to first isolate your particle from the (irrelevant) rest of your image, an operation that is called segmentation and this operation may turn out difficult or even impossible depending on the image structure.

You tell us that you converted your image to 8bit. In case the original image is 16bit or more, don’t convert to 8bit.

Furthermore, please post the original image in its uncompressed format, i.e. as TIF- or PNG-file. JPG-compressed images are unsuited for scientific analyses, because this compression is lossy and introduced artifacts that can’t be removed

Regards

Herbie

Herbie,
Thank you for your quick response.

Here is the original image with the particle in question isolated in TIF format.
0063%20Shale%20x1k%20-%205|nullxnull

Here are the next steps I’ve taken, keeping the image as 16 bit as you mentioned:

Setting threshold:

Applying threshold:

Particle measurements:

Particle analysis:

Results:

Kind regards,
James

James,

thanks for the original image that arrived again as 8bit.

The thing is that you must first get a clear closed binary-valued boundary of your sand grain. Without pre-processing this won’t be possible with simple thresholding!

I’m unable to succeed even with elaborate pre-processing and the reason is the uneven illumination of the grain. Furthermore what you call isolation isn’t perfectly true because there are connected parts that don’t belong to the grain.

If you had a closed binary contour of the grain, which I believe is hardly possible to get automatically, you could use fill holes to remove everything inside, leaving the desired grain for analysis.

I’m sceptic that you will find an automatic means to perform the desired analysis for the sample image.

I fear you have to improve image acquisition, i.e. get much more even illumination and less shadows and less disturbing structures that join the grain of interest.

Good luck

Herbie

Herbie,
Many thanks for your help, I had feared the shadowing in the image may prove difficulty. As a second approach to obtaining the aspect ratio and convexity, I am attempting to manually obtain the values for the following formulae:

Aspect Ratio=dmin/dmax

Convexity=Pconv/P

where dmax is the major axis length (the longest length passing through the centroid of a particle), dmin is the minor axis length (the length of the particle perpendicular to the major axis), P is the perimeter of a particle and Pconv is the perimeter of a particle’s convex hull (the smallest convex shape which contains the particle).

This isn’t proving overly difficult for aspect ratio, however is causing some difficulty for convexity.

In your experience, do you have any recommendations for alternative software to obtain these parameters for the particle image in question?

Kind regards,

James

James,

please try to understand that we aren’t speaking about software but about mathematics and physics.

If you just try to understand what happens in your essays you should be able to logically understand what’s the basic problem. Of course you get a lot of particles and it is absolutely clear why this is so. If it is still unclear to you, then please think purely logically. The problem is not linked to any kind of software but to the property of your image or the objects therein.

In your experience, do you have any recommendations for alternative software to obtain these parameters for the particle image in question?

There are highly advanced methods but I doubt they will help you much further with images of the kind you have provided. One such approach would be WEKA which is a pixel-based classifier that can be used for segmentation. But be aware of the fact that the more involved the methods, the more involved is their correct use!

As long as you have trouble to perfectly understand why your approaches fail, I’m not sure if I can full-heartedly recommend more involved methods.

If you are willing to follow a manual approach and draw a selection around your grain, then you can easily make all kinds of measurements ImageJ provides.

Regards

Herbie

Hello Jamesleak,
I don’t know how many images you have and the variety of shapes but this is what I have been able to do and the results.
Jamesleak_cleared%20binary|nullxnull
Results.csv (158 Bytes)
If you need more information then let us know. SEM and TEM images are typically on a scale difficult to determine.
Bob

It looks as if the top-right part of the grain was manually isolated from the surrounding structures.

Regards

Herbie

Hi Herb,

It was, that is why I asked how many and variety they have to deal with. It was not ment to be perfect but thanks for asking.

Bob