Segmentation and thresholding of reflective droplets

imagej
segmentation

#1

Hi everyone! I would like to ask help in figuring out a way to properly segment this image. I’m very new to imagej and have followed tutorials and read notes on some plugins. However i still cant get a good segmented image after several tries.

Here’s the image i an referring to.


I’m tracking the growth of water droplets and measuring them as they grow. I need to first be able to segment the image prior to analysis.
However, no matter what process i take, i cant seem to isolate the droplets from the background because of the poor contrast.
Here’s one of the images i got after some thresholding and hole filling:

This seems like a very simple problem compared to what i’ve in the forum, ive tried watershed and still nothing good comes out :frowning:

Id really appreciate any help on this please.
Thank you very much.


#2

This is the best i got so far :frowning:

image


#3

Maybe you should try a combination of edge detectors and binary operations.

Or you give machine learning tools like ilastik, Weka, Knime or Intellesis (free trial available) a try.


#4

Good day Yzabelle,

besides other problems, could you please tell us what is the reason for the horizontal stripes in the droplets?

To avoid reflections use a highly diffuse illumination.

Regards

Herbie

PS:
You may use dyes to get more contrast.


#5

thanks for your suggestion sebi06, i will try trainable weka on this.


#6

Good day herbie,
The stripes are actually inherent in the surface. I used copper in the picture, and will also be using brass which also has stripes. Unfortunately the stripes are magnified by the droplets. Im using an LED ring light as my light source. ill try to reduce the intensity of the light and see if it will help. I cant use dyes though because the i have to ensure the copper samples are not altered as they are already pre treated.
Thanks for the suggestion herbie!


#7

ll try to reduce the intensity of the light

No, do not reduce but use a strong diffusor (ground-glass or matt screen). Led light is highly directed and unsuited without diffusor.

You may also try to shine the light on a white screen and use the diffuse reflected light for illumination. However this may lead to uneven lighting. It is really worthwhile experimenting a bit with the lighting.

Regards

Herbie


#8

@Yzabelle

For sure TWS is a great place to start… another good option to check out would be the MorphoLibJ plugin. It’s another great tool for segmentation that comes with Fiji, providing more options/techniques… it comes with Morphological Segmentation tool. Perhaps with a bit of pre-processing… you’ll do fine with either of those options.

You can always test things out and post again with more questions. We are here to help.

eta


#9

Hi everyone! here’s an improvement in the results of what im doing :slight_smile: I played around with the lighting and this current set up seems to be the best option ive tried so far :slight_smile: Thanks Herbie for the suggestion.

here is the result of thresholding:
image

And here is the mask used in counting the droplets and measuring the area:
image

Still needs a lot of improvement though slight_smile: