差分度量的三维模型语义部件定义方法  

Definition Method for Semantic 3D Model Segmentation Based on Difference

作　　者：谭小慧 朱晓 贺振轩 杨洪鼎 李俊 TAN Xiaohui;ZHU Xiao;HE Zhenxuan;YANG Hongding;LI Jun(Capital Normal University,Institute of Artificial Intelligence Education,Beijing 100048,China;Capital Normal University,School of Information Engineering,Beijing 100048,China;University of Exeter,Department of Computer Science,Exeter,U.K.,EX44QJ)

机构地区：[1]首都师范大学人工智能教育研究院,北京100048 [2]首都师范大学信息工程学院,北京100048 [3]埃克塞特大学计算机科学学院,埃克塞特EX44QJ

出　　处：《小型微型计算机系统》2024年第12期3050-3056,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62272324)资助;北京市自然基金项目(4222023)资助;首都师范大学“人工智能赋能首都教育改革与发展”科研项目(RGZNJY2023-YB-16)资助。

摘　　要：本文提出了一种基于差分度量的三维模型语义部件描述符计算方法.当三维模型发生缺失变化时,模型上点的全局描述符产生相应的变化,利用这两者变化的映射关系,使用随机森林算法进行描述符变化量到对应部件的标签映射,可实现对部件的语义特征提取.对于一族模型的标签化问题,本文利用提出的部件描述符计算方法,将模型中对称的部件统一化标签,解决了模型部件的对称部件快速标签化问题.使用半监督的方法,解决了族内模型的差异性问题.在公开COSEG数据集上进行模型部件自动语义标签化,实验结果证明了所提方法的有效性.In this paper,we propose a method for labeling the semantic segmented parts of 3D model based on descriptor variation.Based on the observation that global descriptor of points on the model changes on semantic the part missing.To establish the mapping relationship between descriptor and certain missing part,a random forest algorithm is used to map descriptor change to the label of the corresponding part.The semantic feature of the part is extraction by the proposed method.For the labeling of a family of models,we solve symmetry model parts labeling problem by component-level descriptor.By semi-supervised method,the variance problem of intra-family models is solved.Automatic semantic labeling of model part is carried out on open COSEG data set.Experimental results show that the proposed method is effective.

关 键 词：三维模型 模型分割 描述符 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]