I have two images with severeal non-convex areas. I need to find vector in which I can move area from image 1 to position in image 2.
Problem is, that areas change shape and also can vreak into multiple smaller areas (this is not problem, I use Morphological operator to enclose areas together).
After some operation (median filter -> tressholding -> binary image -> morphological enclose) I have final areas without noise and "errors"

For. example:
a) on first image i have square, but on second image it could be circle
b) area on edge of image... in one image is only part of square (I part of L), on second image is full L shaped area

I have identified corresponding areas in both images. That I use convex hull and calculate its centroid for both areas and create vector from those centroids... problem is, that its not universal solution. Succes rate is about 60%.
Can aynone help me with some suitable solution ?

Concrete example in image:
Old:
[attachment=9832:xxx1.bmp]
Actual:
[attachment=9831:xxx0.bmp]