Hierarchical Surface Abstraction Using Adaptive Mean Shift

This paper presents a novel hierarchical surface abstraction technique based on Mean Shift for 3D shape matching. The automatically generated statistical modes and their associated geometric properties from Mean Shift-based surface analysis are further integrated into an Attributed Relational Graph (ARG). Therefore, the ARG can be used as an abstract representation of the 3D surface object for 3D shape matching, which has direct applications on motion synthesis, capturing and transfer. In particular, we propose an adaptive mean shift technique and a hierarchical framework based on normalized cuts. Our experiments show that the surface abstraction technique is robust, insensitive to resolution changes and stable to noise. We have tested the technique in 3D shape matching. The experimental results demonstrate the effectiveness of the presented method.