particle detection with poor background

[Tommy77]

I am trying to measure the area of growing yeast particles in an image sequence of about 100 images. The images have a bad background which is not possible to remove before image analysis. I have included one image. My VI (attached) does not separate background from particles that well. I have tried numerous methods but none have worked so far. I have tried median filters (before edge detection) and smoothing functions (after edge detection) but they did not work as well as I had hoped.
Does anyone have some ideas or possible solutions?

[Tohatsu]

there are mathematical functions called dilate, grow and extrude that might help you, I did something similar a few years ago where I counted the number of green balls in an "ocean of balls"

What I learned from that was that thresholding is very useful!

Try to obtain a b/w image and then so your math on thatone.

There are operations that can close open circles. depending on what you need, maybe a simple threasholding and a heavy dilation of the edges will work.

I did not do my stuff in LV, but the approach would be the same I guess...

 

Best one:

Message Edited by Tohatsu on 03-28-2008 09:41 AM

[AnalogKid2DigitalMan]

Would it be possible to use stain to increase the contrast of the yeast particles?

Is it possible to change the illumination method? It looks like you may be using backlighting, could you try direct illumination or darkfield methods?

Another option- colored filters to increase the contrast?

[Paul_m6]

See the code I Have attached. I used some functions that you can use for this, see the help for more explanation. This is not a solution but probably will help you. Keep in mind that all the parameter in this vi are static, if the characteristic of the images change you'll probably need to change this parameters dynamically. The code find some of the particles, it would be better to use a skeleton operator and a hough transform for detecting circles in the final part, or maybe something different. I have attached the result of the processing too. Good luck.

Mensaje editado por Paul_m6

[Tommy77]

Thanks for the help, especially to Paul_m6. I have tries to skeletonize and I have used the circle/ellipses detect in IMAQ detect shapes. However, the result is not satisfactory and I want to try the hough transform.
Does anyone have a vi with ellipses detection using a hough transform?
I have only been able to find one for straight lines and have been unable to convert it to ellipses detection.