结合深度特征和信息量的三维模型标注方法  

3D MODEL LABELLING IN CONJUNCTION WITH DEPTH FEATURES AND INFORMATION VOLUME

作　　者：樊亚春[1] 周明全[1] 宋毅[2] 柳勇光[1] 

机构地区：[1]北京师范大学信息科学与技术学院,北京100875 [2]中国兵器科学研究院,北京100089

出　　处：《计算机应用与软件》2011年第11期40-44,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61001168)

摘　　要：针对三维模型语义检索应用,提出一种三维模型语义自动标注方法,建立三维模型内容特征和语义特征之间的映射关系。首先,利用基于深度信息的特征提取方法计算三维模型形状特征描述符,在单位立方体的六个面上正交投影后获取六幅深度缓存图像,提取图像二维傅立叶变换后的270维低频系数作为三维模型内容特征。其次,针对语义词汇之间相似度计算需要,提出一种语义相似度计算方法,采用本体层次结构的深度、宽度、同义词集密度信息计算词汇信息量,定义语义词汇间的信息量关系,得到语义相似度。再次,利用语义排歧策略消除语义词汇二义性,提高语义词汇相似度计算的准确性。最后,融合三维模型内容特征相似度计算和本体语义相似度计算方法,利用样本库中相似模型包含的词汇概率信息和模型内容相似度值,计算待标注模型的语义描述信息。通过模型标注实验,验证了该方法的准确性。Aiming at applying the semantics retrieval of 3 D model, a method of automatic labelling the 3 D model semantics is proposed in this paper, in which the mapping relation between content feature and semantics feature of the 3D model is built. First, the shape feature descriptor of 3D model is calculated by using depth information-based feature extraction method, and six depth buffering images are obwined after orthogonal projecting on six faces of a unit hexahedron, after extracting image＇ s 2D Fourier transform, its 270 low-frequency coefficients are used as the content feature of 3D model. Secondly, in order to calculate the similarity of semantic vocabulary, a new semantic sim：~ arity algorithm is presented. It uses the information of depth, width and synset density of ontology hierarchy to calculate the vocabulary information volume and to define information volume relations between semantic vocabularies so as to obtain the semantic similarity. Thirdly, semntics disambiguation method is employed to eliminate the ambiguity of semantic vocabulary and to improve the precision of similarity calculation with regard to semantic vocabulary. At last, the semantic description information of the model to be labelled is calculated by fusing two similarity calculation methods of both the content features of 3D model and the ontology semantics and by making use of the vocatmlary probability information contained in similarity model in sample library and of the content feature similarity value of the model. The precision of the method proposed is validated with model labelling experiment.

关 键 词：三维模型 深度特征 词汇信息量 语义相似度 语义标注 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]