NOTICE! This is a static HTML version of a legacy Fiji BugZilla bug.

The Fiji project now uses GitHub Issues for issue tracking.

Please file all new issues there.

Bug 1025 - Stitching in Fiji is not using stage coordinates from CZI correctly
Stitching in Fiji is not using stage coordinates from CZI correctly
Status: RESOLVED INVALID
Product: Fiji
Classification: Unclassified
Component: Plugins
unspecified
PC Windows
: P5 major
Assigned To: ImageJ Bugs Mailing List
Depends on:
Blocks:
 
Reported: 2015-03-13 02:54 CDT by Sebastian Rhode
Modified: 2015-03-13 12:50 CDT
1 user (show)

See Also:


Attachments
Comparison for ZEN and Fiji Stitching results (946.47 KB, image/png)
2015-03-13 02:54 CDT, Sebastian Rhode

Description Sebastian Rhode 2015-03-13 02:54:36 CDT
Created attachment 240
Comparison for ZEN and Fiji Stitching results

Hi,

I just tried to compare the ZEN stitching with the Fiji one, since Fiji Stitching has some very nicefeatures, we are still missing ...

The test data set can be found here:

https://dl.dropboxusercontent.com/u/623476/Overview_10X_DAPI.czi

The problem is, that the Fiji stitching is "moving" some tiles to a position, where they do not belong. And the result has empty tiles, while the original CZI images does not have any empty tiles.

The original area is a rectangle, so the result must also be a rectangular tile area (since this is knwon from the stage coordinates, since I use "Positions from File" & "Defined by image Metadata"). But in Fiji the result looks different --> see attachment.

It seems to me, that Fiji is moving some tiles, where is "not much to stitch" to some XY position instead of leaving them at their original positions, when the tile is rather "empty" (just an idea).

Do you have any idea, how solve this issue?

Sebi
Comment 1 Curtis Rueden 2015-03-13 12:50:47 CDT
Make sure you uncheck the "Compute overlap" option. Otherwise, the stitching plugin tries to optimize the error of the tile overlap. When there are tiles with very low signal, this can get weird, as you have seen.

Alternately, if you need the stitching plugin to tweak the overlap a bit, but not too much (as is seen here), you can reduce the value of the "regression threshold" to 0.05 from the default of 0.30, which is too high for some of this dataset's tiles.

Arguably the behavior could be improved here to somehow guess a better regression threshold in a contextual way -- feel free to file a feature request on GitHub. But I expect that none of the core developers would have time to make things more automagical any time soon. Pull requests are very, very welcome though! :-)