摘要
针对三维模型的分类问题,提出了一种适应性加权非对称AdaBoost隐马尔克夫模型(HMM)分类算法.算法中提出了由三维模型表面的绝对法向量表示的两种新特征,将经过归一化和姿态调整的三维模型划分为若干部分,各部分对应HMM的一个状态,对各部分提取特征并用主成分分析(PCA)降维,对模型的4种特征对应的弱分类器使用非对称AdaBoost算法进行boosting.HMM的结构及参数初始值由模型姿势调整的可能形式及观测顺序确定,训练过程中参数用期望最大化方法计算,最后使用加权相似度计算对三维模型分类.分析及试验结果表明,与基于分布函数的分类算法相比,该算法明显提高了正确率.适应性加权后,分类正确率可进一步提高.
An adaptive-weighted asymmetric AdaBoost hidden Markov models （HMM） classification method was proposed for 3D model classification. Two new types of features embedded in absolute normal direction of surface were proposed to describe 3D models in the method. A model was split into several parts after normalization and pose adjusting, and each part formed a state of a HMM. Features were extracted from each part, and the dimension of features was reduced by principal component analysis （PCA）. Asym metric AdaBoost was introduced to boost weak classifiers corresponding to 4 types of features. The structure and initial parameters of HMMs were determined by different possibilities of pose adjusting and parameters were estimated using expected maximization algorithm. The class label of a test model was determined by using weighted similarity calculations. Analysis and experimental results showed that the method can gain much higher classification accuracy than the classification method based on distribution function. The classification accuracy can be even improved letting weights adapt with test models.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2006年第8期1300-1305,共6页
Journal of Zhejiang University：Engineering Science
基金
国家自然科学基金资助项目(60273056)
国家"863"高技术研究发展计划资助项目(2003AA411021)
