Extracting images, advice needed

Hi Community,

I’m looking for an advice how to approach a process.
The robot is supposed to open a scanned document (different but finite layouts, formats jpg,pdf,png,tiff) and find a place where is a frame with customer handwritten signature. If found the frame area should be cropped, saved as image and uploaded to dedicated application.

The problem is with finding best solution for image analysis and cropping. Each document may have a static anchor which would inicate where the frame with signature is.

I’ve played around with displaying the document in a viewer, finding anchor image matches and taking screenshot of a relative clipping region, but this is not too reliable, and very non-elegant solution.

Could you think of any better way?
Would flexicaptue be of any help, since actually there is no ocr involved?