In this talk I will describe solving of an industrial project of an automatic jewelry stones classification. Classification is process of categorizing a stone to one of the predefined classes. The stones may differ in size, shape and damage type. Originally the project was meant to be only an image fusion. Later stone registration function was added and finally our own classification and machine learning process was implemented. The presentation will demonstrate situation where two different worlds meet: an industrial which is strictly application focused and scientific which is mainly focused on theoretical results. The final solution is a research report of our own abandoned methods which are not now applicable due to speed requirement and implementation of our methods which fully satisfied given requirements, but are not strictly novel.