Classification of images only



I’d like to use only measurement on image basis to run the classifier. Is it possible? If yes, how would I have to set it up?


This is possible in the following way:

  • Add an ObjectNumber column to your per-image table, and set the values to 1.
  • Create a new primary index for the per-image table consisting of the ImageNumber and ObjectNumber columns.
  • Add X-location and Y-location columns to the per-image table; you can set the value to whatever you want, but I’ve used the image center myself.
  • In the properties file, have the image_table
    and object_table fields both refer to the modified per-image table.
    What this basically does is trick CPA into treating the per-image as if it were a per-object table; Classifier will then treat it accordingly. It’s kludgy but it works.



Hi there,

Then, could you tell me how to add X-location and Y-location columns to per-image table?
Can I add these columns in Metadata module in CellProfiler or CellProfiler Analyst?



This is an old post, CellProfiler Analyst now allows you to do per-image classification without having to do the above steps. They aren’t necessary for the Image QC workflow you were working on in the other post.