Curvature-direction measures for 3D feature detection

In this paper, we propose a new robust feature extraction algorithm for 3D models based on principal curvature direction. Generally, the feature regions tend to be more noisy, so it demands a robust technique to handle features effectively. Because the integral invariants are robust against noise, the principal curvature information is estimated based on principal component analysis. After fuzzy filtering of the principal curvature direction, it becomes a good description of the geometric discontinuity. Compared with the curvature values, the impact of noise on the principal curvature direction is small. Therefore, feature extraction based on principal curvature direction is more robust and accurate. The experimental results show that the proposed algorithm can efficiently extract feature and distinguish noise.