Help optimizing acquisition for stitching

Hi Everyone,
I’m hoping to get some input/ideas on how to optimize acquisition parameters for image stitching.
In general, I’ve seen that the rule-of-thumb for tiling and stitching is that you want ~20% overlap of the frames (or at least that’s the common default). The downside of 20% overlap (in a grid) is that this means that 64% of your image data is acquired twice.
If you are doing a lot of imaging, say a screen, this can lead to a lot of time acquiring redundant data.

Does anyone have any experience in optimizing the 2D tiling overlap for maintaining stitching accuracy but minimizing the overlap (i.e. decrease imaging time per plate)?
I’m looking for hints/ideas on how to best to tackle this optimization.
-if it matters, I’m using Zeiss-Zen for imaging, which can stitch the images concurrently with acquisition.


Hi @JTRodgers: just wanted to my two cents on this.

First I would seek to understand how the stitching is performed in your software of preference. This will help to inform your choice of overlap.

If your algorithm uses landmark points in the overlap areas to align areas common to both images, think about how much “texture” there is in the border of your images. A fluorescent image of sparse cells with a fairly uniform noise profile won’t help much unless a good number of cells are in the overlap region. Compare this to a brightfield tissue image which will have lots of unique texture features on which to align.

The second point I would make is to be sure to check that you’re getting uniform illumination on your field. Even if it looks good to you, you may be able to identify problems when you enhance the contrast.
This can cause a real problem for stitching as the overlap part of your field will likely not look the same in adjacent tiles.
This example comes from the end of this article you might find helpful:

Overall, I would say it’s worth acquiring a small test data set of images with a range of different overlaps and see how well it performs using different settings (and software) until you find the minimum overlap that reliably works for you.

Hope that helps!