Extraction d'attributs et mesures de similarité basées sur la forme

TL;DRAbstract

For object recognition and similarity retrieval, object shape features are powerful since they are strongly linked to the object. This thesis develops three new shape descriptors: scale-controlled area-difference shape descriptor (SCAD), normalized partial area vector (NPAV) and chord context descriptor. SCAD and NPAV are based on iso-area normalization (IAN). SCAD is a vector whose elements are the area differences between the main corresponding segments of original contour and its filtered version. NPAV is an affine invariant descriptor based on two proposed theorems taking partial areas into consideration. Both SCAD and NPAV have the good property of affine invariance. The SCAD is more compact than the NPAV but the performance of NPAV is much better than that of SCAD. Chord context describes a frequency distribution of chord lengths with different orientations. It is shown to be unaffected by shape translation, rotation and scaling; it is also robust to minor occultation, non-rigid

Chat with Paper

AI Agents for this Paper