Counting Beans

Image Analysis
#1

Hi everyone,

thanks for this great tool, if I only knew how to use it.

I would like to use it to count beans and measure the surface, but I came across a few problems putting together my pipeline.

  1. seeds lying closely together are identified as one big object (as shown in attached picture
    results summary.jpg960×720 30.7 KB

    ). I’ve been playing with all options I could find, but without much success, I’m not sure if it is a problem with the threshold or with distinguishing clumped objects.
  2. the variation in color is proving to be difficult, especially white beans are often not captured. I thought about making two analyses of different images of the split color-to -greyscale results. is there a merge function to combine these two analyses for object identification?

I hope someone has a good idea for me.
Thanks a lot,

the beancounter

#2

Hi beancounter,
Would you mind posting a couple of example images, so we could take a look?
Regards,
-Mark

#3

I thought I did.
Or do you mean the original fotos?

here are two:

light flash 20140302_131907.jpg2560×1920 600 KB

light flash 20140302_131907.jpg2560×1920 600 KB

the edge enhancement function actually captures the shapes very well. But I didn’t find a way to use this for the object calling.

light flash20140302_131641.jpg2560×1920 616 KB

#4

Hi,

This is not an easy problem, at least for what CellProfiler was designed for. Please see my attached pipeline, at least for some ideas on how to attack this.

  • Crop the wasted pixels if at all possible, or at least reduce the resolution. CP may run into memory errors otherwise in any filtering step.
  • Unmix Colors is meant for histological images, but seems to do ok here. You might also try the Methyl green setting.
  • I tried EnhanceOrSuppressFeatures, but you could remove this, as the EnhanceEdges seemed to work better. Either of these has the added benefit of removing the substantial background illumination artifact.
  • IdentifyPrimaryObjects - You will need to adjust the declumping parameters.

Better, though, would be to use a black background for your images if at all possible. Our thresholding assumes that the intensity of the grayscale foreground will be higher than the background. Also, there is a substantial illumination variation, which you can alleviate by CorrectIllumination[Calculate/Apply], but it’s better to minimize it depending on your light source.

Best,
David
DLpipeline.cppipe (7.27 KB)