I have a problem of images not being stitched very well (please see image). Images are taking by Zeiss Axio imager and stitching is done in Zen blue (3.6).
I used to apply the same stitching parameters and never had this issue, and I have no idea what could be the reason. I played around with stitching parameters (minimal overlap, max shift etc) but nothing fixed it. The usual stitching parameters when taking tile images with the advanced settings are: tile overlap: 10%, when stitching: minimal overlap 5%, max shift 10%, Optimized. Worth to note that some areas look nicely stitched, while some others are completely missed. Please let me know if you need more information. Thank you very much for your input.
I tried it already and it fixes some tiles but ruins the stitching somewhere else in the image. I used to stitch images with more than 150 tiles and they were perfectly stitched. This issue is new and nothing in the microscope or stitching settings changed!
Here’s the thing about correlation thresholds in stitching - they highly depend on the level of contrast and detail in the overlapping regions. Sometimes seams line up with areas of the section that have minimal detail, or even completely black areas. Software will always have a tough time with these areas - so minimizing them is key. It sounds like you’ve been lucky so far - until now.
10% overlap is an absolute bare minimum in my experience - we typically recommend 20% for most tissue at 10-20x mag. Generally the lower the mag of the objective the more overlap you want to acquire. There are certain types of tissue that benefit from >30% overlap with low mags like a 10x.
Overexposure will also negatively influence any seam correlation so watch out for this (the image above looks like it has quite a bit).
You don’t mention the objective you are using. For best performance use an optic that has a ‘Plan’ correction - which means it has a field flatness correction. Objectives without this correction often exhibit severe pincushion or barrel distortion that makes stitching incredibly difficult unless the software can correct for it. Most scientific software like ZEN does not warp the image to correct for these types of artifacts - so you need to choose objectives wisely when depending on ZEN for stitching.
When we do processing in ZEN - we like to set out minimum overlaps to match the actual overlap set in the acquisition. Max shifts can be equal to or slightly greater than this - but this will have to be determined empirically as each section is different and some may benefit from including a larger search area. But as you have noted - it can also create problems elsewhere.
Finally, set your Comparer and Global Optimizer to Best and use the Edge Detector option.
This could be indeed something related to the software but just for the sake of exploring all options there could be another reason. I have seen similar problems at the facility when users do not properly fasten their slides to the sample holder. Due to the fast motion of the stage and inertia it can happen that the slide moves a tiny bit while the tile experiment is carried out. This happens usually in a step fashion and “randomly”. Due to this, after stitching the output image looks very good in some areas but these larger areas can not be stitch together.
Thank you very much everyone for your input.
Rafa you´re right I´ve seen this problem before also. However, this is not the case in my slides. They are perfectly held on the stage. Jason I´m using Plan-Apochromat 20x/0.80 Ph 2 M27 (but the phase contrast is not inserted while acquiring pictures). I will try your suggestions and see if the stitching is better! What puzzles me still is that worked fine for me for more than a year, and now it´s not the case. However I have to mention that I used to work with the axio 503 color cam, and now I switched to axio 503 mono. I don´t know if this could affect stitching?
The change in camera could be relevant. Did you check that for the mono cam the rotation is well corrected? basically that when you move the stage horizontally, so does the features in the image? I have seen some problems with the stitching algorithm if it also has to compensate for a (large) rotation. Sorry for the vague term but I don’t think I ever checked at what angles things go bad.
so I applied 20% overlap as you suggested and max shift 22%, comparer and global optimizer to best and edge detector, it looks perfectly stitched! Thank you very much for these information! Great help!
For stitching we always use the smallest possible minimal overlap (I think one percent in Zen) and at least 15-20% for maximum, and also choose the most detailed channel and z-plane as reference.
Additionally, a possible reason for sudden drop in stitching quality may be misalignment of the camera - if someone tinkered with it or pushed it unwillingly and rotated it, then the lateral overlaps become variable because the stage now moves obliquely related to the camera. So try realigning your camera to the x and y direction of your stage.