I was examining the error for a while... Sorry for the delay.
The example you show here "multi-channel-time-series.tif" actually has an issue.
Although it could be loaded "fine" in ImageJ, but when you open "Image" >> "Properties"
You can see that:
Channel (c) : 1
Slices (z) : 21
Frames (t): 1
Which is the exact reason why when you load into CellProfiler, you would have 21 rows and only 1 channel.
I agree that the metadata written in xml is written with 3 channels, but the properties is written incorrectly.
A quick fix in the Properties, set it to:
Channel (c) : 3
Slices (z) : 1
Frames (t): 7
Then save the image >> Restart CellProfiler and load the new image with correct properties.
Now you will have correct time/slice/channel when using "Extract from image file headers" as you did.
note: Python indexing starts from 0.
Back to your own data, can you try open your image with ImageJ and make sure both metadata in xml and in properties agree to same values channels/slices/time. Then save the images and open in CellProfiler.
Hope that helps you both.