background subtraction variability

I am experiencing a problem with background correction that seems to be common. Perhaps someone in the forum group has already solved it.

I did 2 background scans this morning, and then 3 replicate scans of the small optical density discs within 10 minutes. When I later batch processed these images, the first 2 replicates had good background correction and correctly detected 54 discs, but the third replicate had a problem with two dark areas in the processed image and thus incorrectly drew lines around more than one disc (in one case) within one of these dark areas, and perhaps no lines around some other discs also within the dark areas. As a result, it identified only 50 discs. In other words, it missed counting some of the discs in the dark areas. A few hours later I repeated a scan of the discs (replicate #4; I had not disturbed the sample inside the zooscan in the meantime) and then processed this image in batch mode using the background image from the morning. Replicate 4 had excellent background correction and correctly identified 54 discs, even though the background image was separated from replicate #4 by 5 hours, and from replicate #3 by < 1 hour. So, the question is, what happened during the processing of replicate #3? I did not touch the sample in the zooscan all day, i.e. between replicates 1 and 4. I did not change any settings on any of the menus. How can the background correction go wrong on only one of a series of replicate scans? More to the point, if this happens following overnight batch processing of an entire day's worth of zooplankton samples, what can I do short of having to scan the improperly corrected samples again? Can I reprocess the raw image using the closest background sample over again? Will this help? Or, might I need to reprocess the background images from the previous day first, and then reprocess the raw image? What do you advise?

Don