how to shape bounding box and get correct transform matrices for multiple images

Question

actually OpenCV can do whatever we want, however it is a fundamental library with merely low-level APIs.

my question is how to calculate transform matrices for multip