Scale and Orientation Invariant 3D Interest Point Extraction Using HK Curvatures

Abstract

Although they are orientation invariant, mean (H) and Gaussian (K) curvature values are essentially variant under scale and resolution changes. In order to overcome this fact, in this study, scale-spaces of the 3D surface and the curvature values are constructed. Then features with their scale information are sought within the scale-space. Thus, different from previous studies, H and K curvature values are obtained using constant threshold values and independent of the scale of the 3D surface. Also, 3D features are extracted with their sizes over the surface. Consequently, salient features extracted from a 3D surface are comparable to their identical but resized versions found on the scaled versions of the same 3D surface. In other words, metric sizes for each feature found over the surface are given and by this way complete scale and resolution invariance is assured. Moreover, robustness of feature extraction under scale and noise is tested. Also, the method is used for object recognition when a database is constructed by virtually resizing the Stuttgart database objects. The results are compared with the ones obtained when scale space is not used.