This paper presents new concepts in the design and implementation of a virtual endoscope. Starting from standard magnetic resonance images, and applying a two-step image processing, inner views of the human body can be obtained, even of such parts of the body, which cannot be penetrated by a traditional endoscope. The first step of the image processing consists in an enhanced version of the fuzzy C-means segmentation. Then a shape recovery algorithm is employed in order to reconstruct the 3-D object. The algorithms provide good-quality segmented images a very quick way, which makes them excellent tools to support a virtual endoscopy.