Image matching from an existing C# application

[juergenboehm]

In a machine controlling software we have the problem of matching two images A and B with each other, where both are photographs of a part of a PCB which are taken with a camera from inside the machine. Lighting can vary and A is sometimes a part of B, sometimes nearly identical in size and partial occlusions can occur.

Currently we simply try to estimate a translation vector that makes A align with B. For this we use a simple minded cross-correlation implemented on top of an FFT package for our C# machine control app. But the results are problematic, often no translation vector can be singled out reliably from the cross-correlation data and occlusion and lighting pose additional problems.

So we started to look for a modern, state-of-the art solution and after some googling I found this software - with the WebInterface on HuggingFace Spaces we tried some of our A/B image samples, but now I am a bit stuck what to do next:

1. Is it possible to write a short (local) python script that takes as input two image files A and B and computes a transformation of the plane (most generally a projective transformation from PGL(3,R)) that matches A and B using somehow the web interface of Image Match WebUI? (We have at the moment very little or no experience in API-centered Python or Javascript-based programming, but develop mostly standalone WPF applications in C# and with C++ extension dlls).
   How could such a script look like?

2. Same question as 1. but could it be done from inside a C# code, somehow connecting to a web service running on a local machine (localhost or in company net)?

3. If using the Image Matching app over the web is too much overhead and most of its functionality (which is provided as a survey and for experiments) is not needed anyhow what would be the steps to isolate a specific method from it and make it available for use from inside our C# code?

Many thanks for any answers!

Jürgen