Annotation of endoscopic videos on mobile devices: a bottom-up approach