Abstract: Non-rigid 3D objects have plenty of shape deformations because of posture variations, so non-rigid shape retrieval is more challenging than rigid shape retrieval. Shape descriptor is especially important to non-rigid shape retrieval. In order to improve the retrieval accuracy, a new global shape descriptor for non-rigid 3D object is proposed in this paper. The key idea of the approach is to represent the SIHKS（Scale Invariant Heat Kernel Signature） local shape descriptors by means of the sparse representation theory, so it is called SR-SIHKS. The computation of SIHKS is improved by adaptively deducing the time parameters from the non-rigid benchmark. K-SVD algorithm is adopted to train a dictionary, and the sparse representations of local shape descriptors are gained by Batch-OMP algorithm. The sparse representations of all local shape descriptors are integrated over the entire shape to form a global shape descriptor. Experimental results show SR-SIHKS has obviously better retrieval performance than SIHKS and HKS on some non-rigid shape retrieval benchmarks.

Key words: 3D model retrieval, non-rigid object, shape descriptor；heat kernel signature, sparse representation

摘要： 非刚体由于姿态变化会产出多样的形变，因此非刚体的形状检索比刚体更具挑战性。形状特征提取是非刚体三维模型形状检索的关键问题。为了提高非刚体形状检索的准确度，提出了一种非刚体全局形状特征提取方法。此方法的核心思想是将稀疏表示（Sparse Representation，SR）理论用于对尺度无关的热核特征（Scale Invariant Heat Kernel Signature，SIHKS）进行稀疏编码，因此被称为SR-SIHKS。改进了SIHKS局部特征的提取方法，根据所处理的模型库来自适应地确定热扩散时间参数；采用K-SVD算法来训练字典，借助Batch-OMP算法实现局部特征的稀疏编码；将非刚体三维模型的所有局部特征的稀疏编码汇聚为全局形状特征。实验结果表明，SR-SIHKS具有比SIHKS和HKS更优的检索效果。

关键词: 三维模型检索, 非刚体, 形状特征, 热核特征, 稀疏表示