Fig. 1 of Object-centered image stitching shows an typical issue when working with image stitching: a specific part of an image should be preserved but the seam cuts right through it.
I know that seam carving (e.g. here and here) use energy maps to calculate a path where it's "cheap" to cut through. To mark areas which should be preserved, we could penalize cutting though there by increasing the energy in this area.
To me it seams like OpenCVs GraphCutSeamFinder provides something in this direction: bad_region_penalty
However, the find function does not seem to allow making use of something which might penalize a bad region. The masks which are passed in in stitching_detailed are only updating the warped masks.
Can I specify areas which should be preserved during OpenCV image stitching using GraphCutSeamFinder or other seam finders? Or does the bad_region_penalty parameter exists for a complete other use case? The seam finders are implemented here but I'm not familiar with C++