Background Problem



I’ve been working with the yeast pipeline, and after a rough start, I’ve gotten it to process my images without any problems except one- the program is counting very tiny light spots (dust, the area around any writing, the edges, etc) as colonies. Thus, on a plate that clearly only has around 5 or 6 colonies, is being counted as having anywhere from 200- 2000 colonies. The closest I’ve been able to get it, by playing with the image quality and the type of surface I use for a background, has been in the mid 80s. I’m thinking that I need to add something like the Filter by Object Measurement module, so that I can give it a minimum and maximum colony size to prevent it from counting background objects.

Pictures are attached below.

So I am asking, first, if this would be a correct approach, and if not, what should I do? Second, if this module is what I need, I need to know how to go about inserting it. I’ve gone through the manual a few times, and I don’t really have much experience yet with this program, so I’m quite lost on how to do that.

Please let me know,
Thank you,


Hi Rebecca,

It is likely that some altering settings in your existing segmentation module(s) (e.g. Threshold Correction Factor in IdentifyPrimAutomatic) would fix the problem. Can you send your pipeline and an original image(s) necessary for the pipeline for us to take a look at?



One option you might want to try is to use either the Background or RobustBackground methods for thresholding in IdentifyPrimAuto. Both of these tend to work better in the cases where there a few or no objects in the foreground. But as David said, seeing the pipelines and images for ourselves will be most helpful in diagnosing the issue.




Sorry for not replying quickly,
Please find my pipeline, with my template and two original images attached.

Thank you for your assistance,
Tutino Yeast (508 KB)


Hi Rebecca,

The problem seems to lie in that the parameter settings were not appropriate for your images. For example, one to look at is the size-dependent value in CorrectIllumination_Calculate (“Enter the block size…”). Since the images in the ExampleYeastColonies pipeline are three times larger than yours, the values for you need to be three times smaller. Changing this value from 22 down to 7 is a good start.

Also, it seems that RobustBackground will give you a better segmentation in IdentifyPrimAutomatic. However, you’ll notice that the threshold is still ends up being too high, so you will want to adjust the Threshold Correction Factor as David suggested (perhaps from 1 to 2).

Also, you have a piece of what looks like tape in the upper half of the dish; this ends up getting identified as an object in IdentifyPrimAutomatic and will probably need additional filtering to remove.

A key point to remember is that the example pipelines and images are just that: examples. Your images may share many of the characteristics of the example images but may not necessarily be identical. In order to make these pipelines work on your images, you may need to adjust the settings as needed.

For example, in IdentifyPrimAutomatic, the red image is used because the colonies are red in the image and therefore gives the best contrast for colony identification. Your colonies are gray in color and so the red image may not be the best choice. Using CropCombined instead of CropRedPlate in IdentifyPrimAutomatic will probably give you a better colony identification.

Hope this helps!