3D-SUSAN: A Robust Extension of 2D SUSAN Operator for 3D Interest Points Detection

Three dimension model is widely used in industrial design, cultural relic protection, education, animation and entertainment. Many 3D interest point detection algorithms have been developed in recent years. In this paper, we proposes a new 3D interest point detection inspired by the idea of 2D SUSAN operator. Our main contributions is proposing a robust USAN volume estimation method based on the mean curvature analysis of vertices. More specifically, a 3D sphere neighborhood is defined firstly, and an adaptive threshold based on the standard deviation of the mean curvatures is used for USAN volume estimation. Then a set of candidate interest points are extracted according to the 3D-SUSAN responses. Finally, the ambiguous or pseudo interest points are removed by the proposed pseudo interest point elimination and the non-maximum suppression. Visual comparison results and the quantitative comparison results demonstrate the effectiveness and robustness of our 3D-SUSAN.