I found an article (see link attached) which describes a method how to decrease oversegmentation while doing watershed. The method proposed really works for 8bit images, but I cannot implement it for 16bit ones, since I do not understand constants and even where the principle comes from.
Thank you in advance, if you help me on the question or recommend a likely method
http://www.csjournals.com/IJCSC/PDF1-1/34.pdf