The method involves selecting a number of potential candidate objects based on a three dimensional candidate generator from an image data, where the generator has a curvature tensor. The number of candidate objects is classified to produce a set of anatomical objects with a reduced wrong positive rate. The image data are seized from a data acquisition device (34) that has a computer tomography system. Independent claims are also included for the following: (A) a computer medium for recognition of anatomical object (B) a system for recognition of an anatomical object.