一种基于邻域扩展聚类的去噪算法  

Denoising Algorithm Based on Neighborhood Expansion Clustering

作　　者：李幸刚 张亚萍[1] 杨雨薇[1] 

机构地区：[1]云南师范大学信息学院,云南昆明650500

出　　处：《系统仿真学报》2017年第11期2663-2669,2677,共8页Journal of System Simulation

基　　金：国家自然科学基金(61262070;61462097)

摘　　要：针对基于二维图像重建出的带有离群点和噪声的三维点云模型,提出了一种基于邻域扩展聚类的去噪算法。通过数据点之间的欧氏距离以及相邻位置关系的可传递性,搜索每个数据点的邻域,然后对所有点进行聚类划分,从而检测和滤除点云模型中的离群点。重点讨论了点云邻域扩展聚类的概念和方法、如何利用基于动态网格划分法快速搜索点的邻域。解决了点云模型周围孤立及密集分布的离群点检测和滤除问题,提高了传统k-近邻等算法对于点云数据的去噪效率。仿真实验结果表明,该算法可有效滤除点云模型中孤立和密集分布的离群点。Aimed at the three dimensional point clouds model with outliers and noises which is reconstructed based on 2D images, a new denoising algorithm based on neighborhood expansion clustering is proposed. The search for other neighboring points of each data point is conducted by using the Euclidean distance between data points and the transitive property of the neighborhood location relation. All points are processed for cluster partition, which detects and filters the outliers in the point clouds model. The concept of neighborhood expansion clustering and the fast search algorithm based on dynamic grids division are discussed. It solves the problem of detecting and filtering the outliers distributed in isolation or densely around the point clouds model, which improves the efficiency of traditional k-nearest neighbor algorithm to denoise the point clouds data. Simulation results show that the proposed algorithm can effectively filter the outliers distributed in isolation or densely around the point clouds model.

关 键 词：邻域扩展 离群点 去噪 K-近邻 欧氏距离 

分 类 号：TP391.72[自动化与计算机技术—计算机应用技术]