用于快速形状匹配的精确型高度函数特征描述  

Feature description of exact height function used in fast shape retrieval

作　　者：孙国栋[1] 张杨[1] 李萍[1] 梅术正 赵大兴[1] 

机构地区：[1]湖北工业大学机械工程学院,湖北武汉430068

出　　处：《光学精密工程》2017年第1期224-235,共12页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.51675166;No.51205115)

摘　　要：在形状匹配过程中为了提升高度函数描述子的检索精度和对边界噪声与局部变形的鲁棒性,本文提出了一种精确型高度函数特征描述算法。首先提取目标形状外轮廓,构造轮廓采样点的精确型高度函数描述子并进行特征降维,接着利用优化后的并行动态规划进行形状匹配,最后引入形状复杂度分析提升匹配效果。基于点的几何特征显著性,提出形状精度理论,进一步分析局部形变与边缘噪声对形状特征描述的影响。在MPEG-7数据库、Swedish Leaf数据库、Tools数据库和ETH-80大型3D数据库上进行匹配实验以及在Kimia99数据库上进行抗噪实验,实验结果表明:本文提出的算法效率高,匹配时间仅为高度函数描述子的12.5%,在MPEG-7和ETH-80上的检索率最高分别为90.38%和90.07%;在Swedish Leaf和Tools上,检索精度最高分别为95.07%和94.86%,检索性能和鲁棒性均优于高度函数和其他重要算法;在添加噪声的Kimia 99上,该算法的抗噪性能优于高度函数描述子,即使在噪声水平为2.0的情况下,依旧能保持91.92%的检索率。本文提出的算法检索精度高,效率高,鲁棒性好,抗噪性强,具有较好的可扩展性,能有效地应用于形状检索领域。In order to improve the discrimination ability and robustness of contour noise and deformation of Height Functions （HF） descriptor in the process of shape retrieval,a feature description algorithm of exact height functions is proposed in shape retrieval.Firstly,contour outside the target shape is extracted,and then exact height functions type descriptors of sampling points are constructed for dimensionality reduction.And then,the optimized parallel dynamic programming algorithm is employed in matching stage.Finally,shape complexity analysis is used to improve matching effect.Based on point geometric feature saliency,the shape precision theory is proposed to further analyze the influence of the local deformation and the edge noise on shape feature description.The matching experiment has been conducted on the database of MPEG-7,Swedish Leaf,Tools,ETH-80 and noise experiment has been conducted on Kimia99 database.Experimental results indicate that the proposed algorithm in this paper is highly efficient and the matching time of it is only 12.5％ of the original HF descriptor.The highest retrieval ratio can reach 90.38％ on MPEG-7,90.07％ on ETH-80,95.07％ on Swedish Leaf and 94.86％ on Tools respectively and retrieval performance and robustness are better than HF and other important algorithms;on Kimia99 with adding noise,the anti-noise performance of the proposed algorithm is superior to the original HF descriptor,and even in the case of noise level of 2.0,the algorithm can still keep a retrieval rate of 91.92％.The proposed algorithm,with high accuracy,high efficiency,great robustness and noise immunity and good scalability,can be effectively applied to shape retrieval field.

关 键 词：形状匹配 精确型高度函数 形状精度 并行动态规划 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]