摘要
网格显著性是三维网格模型的一个重要几何属性,已应用于许多方向。受现有算法的启发,提出了一种基于局部区块曲率熵的多尺度显著性检测算法。针对每个顶点,定义一个局部坐标系并计算该点曲率值;通过邻域累积体积定义一个改进的自适应区块,计算该点邻域球内每个邻居点相对于该区块的偏离值,将该值作为相应邻居点曲率的加权值;将所有邻居点的加权曲率熵作为该点的显著性值。该算法在时间复杂度方面具有可比较性,在显著性检测能力上占有优势。
Mesh saliency is an important geometrical characteristic of 3D mesh model and has been applied in many applications. Inspired by the existing algorithms, a novel multi-scale saliency detection method based on local patch weighted curvature entropy was proposed. A local coordinate system and curvature value of each vertex was estimated. An improved adaptive patch was defined on the tangent plane using accumulated volume of neighborhood. Furthermore, deviation of the patch of each vertex to their neighborhood was defined as the weight of curvature value. The Shannon entropy of weighted curvature values of neighbor vertices within a sphere centered at each vertex was defined as their saliency scores. Comparisons with state-of-the-art methods have shown the competitive performance in computation speed and the advantage in saliency detection ability of our method.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2017年第9期1976-1983,共8页
Journal of System Simulation
基金
Northwestern Polytechnical University doctoral dissertation Innovation Fund(CX201701)