Particle Analysis

[Steve.Briggs]

Hello all,

When using the IMAQ Particle Analysis vi I take the Bounding Rectangles with respect to the Center of Mass X on a Defined, Thresholded, Masked image.  The result is many bounded rectangles that I convert to ROIs for analysis of the area on each particle.  However, I run into trouble with the amount of data actually being taken into consideration.  The thresholding method I have at this point works great in finding discrete scratches and digs in prisms that I image, but it also takes what can be none other than dust particles into consideration and they are bounded in rectangles that are 1 x 1, 2 x 2, or 3 x 3 (along those lines, very small ROIs).

I'm able to sort through the arrays and separate the needed ROIs from the unnecessary ROIs, but that is only in numerical arrays.  I need to be able to ignore the ROIs that come up on the actual image display (as shown below), but to do that I have to be able to completely throw out certain data in a long array of bounding rectangles.  If anyone has any ideas or questions that can lead to any solutions I'd be very appreciative.

**image is a bit too large to display in post, see attachment**

Message Edited by Steve.Briggs on 11-07-2007 11:07 AM

Message Edited by Steve.Briggs on 11-07-2007 11:07 AM

[Steve.Briggs]

Posted this in the wrong section. 

Anyone kind enough to move it to the proper area (Machine Vision)?

Message Edited by Steve.Briggs on 11-08-2007 08:39 AM

[chilly_charly]

Why don't you filter first your particles, for instance using a minimum treshold on the particle area, or using a shape indicator (since obviously a scratch is far rom circular) ? You could use the IMAQ particle filter 2, in the Vision/image processing/Morphology subpalette.

[Steve.Briggs]

Hey Charly,

Shape indicator would solve the problem if I were only looking at scratches, but I have to take into consideration the digs as well (round and circular).

The current threshold I use is background correction on bright object in my image.  Would you recommend using the Niblack method, and if so, at what deviation factor?

In terms of the IMAQ particle filter 2, the filter is only in terms of pixel value, unless I'm terribly mistaken.  If I filter out a minor scratch or dig in this filter because the pixel value is constantly changing from image to image and prism to prism, then my test is no longer valid.  Does the filter have an option for removing particles by size, rather than pixel value?

[chilly_charly]

I can't help you on the choice of the best threshold method. That's really depend on the images you have top analyse. There is no definitive answer to such a question !

Otherwise, yes, I believe you are terribly mistaking. 🙂 There are a lot of parametersthat can be used either in pixel or in real-world (I suppose this is what you call size ?) dimensions. Don't you have the Vision Assistant to prototype your method ?

Message Edité par chilly charly le 11-08-2007 03:17 PM

[Steve.Briggs]

Currently the background correction method for bright objects gives me the threshold quality I need for the grading of these scratches and digs.

I do have Vision Assistant 8.0 but was never formally introduced to it and have yet to use it .  I don't really know what it's capable of , I assume simplifying the vision process in LabVIEW?

Thank you for your help Charly, I really appreciate it.

[chilly_charly]

Since the Vision Assistant has been available, I have NEVER started the development of a vision application without using it.

Among NI applications, this is really the most impressive tool I have ever used !

You can try any function, evaluate different strategies, prototype fairly complex operations, and when you have finished, generate automaticaly the corresponding LabVIEW vi !

Amazing you don't know that tool. Sure you'll regret it now !

[jrpe]

Hi Steve, if you are doing image analysis, Vision Assistant is the best place to start. It lets you try out all the VIs and actually see what effect they will have. I am actually currently working on a project using vision to track haidinger fringe patterns (2D interference patterns). Funny enough, I am probably going to be working on a program to analyze scratch and dig next! I had zero image analysis experience going into the project, and just playing around with vision I was able to get a working program, its pretty easy to figure out.

You load up an image, and then you can test out what operations you want to do to your image, create a script, and it will turn that into a VI for you. Once you have that VI you can optimize it even more. There are examples in Vision Assistant that show how to use it as well. At least one of the examples relates to particle analysis, I think it is relating to cells.

Anyway, I have found that Vision assistant is a very useful tool for image analysis, you should check it out and play around with it for a few minutes, hopefully it will save you more than those few minutes in the long run!

[Steve.Briggs]

I'll check Vision Assistant out..

I currently have a fully operational VI in LabVIEW so it's not like I'm completely foreign to the idea and ways of Vision.  I just needed some clarification on a few things that are unknown to me for the automation of my ROIs.  I'm not sure that Vision Assistant will really help me with what I'm looking for but like I said...I'll give it a shot.

I'm feeling a bit optimistic today anyway

[Steve.Briggs]

Optimism pays off!

I was able to reduce the automation of ROIs on the smallest particles by almost 72 percent.

Thank you very much Charly for pointing out the usefulness of this tool.