Thank you for this evaluation
I have improve “shape detection” - this term include everything that contributes intuitively or not to what the user wants to do - and in this sense it’s very different from a “clipping technique” (mask, layers…) of the objects we want to treat.
I begin this improvment in January 2019, and it is based essentially on the deltaE, but also on the variation of structure. In most cases this technique should solve the problems.
Over the last few weeks I have added a series of algebra control for deltaE, transitions, … I think today the algorithm is sensibly finalized - even if of course alongside this algorithm there are improvements that can be made over time (regardless of the rigor of the code or its speed);
Ingo is right, before “merge with dev”, we have to check the code, its robustness, improve its speed, and for the GUI part (thanks to Pandagrapher) test the current interface