Abstract-Semantic based 3D model retrieval (SB3DMR) has
attracted more and more research interests, and is a challenge
research problem in the field of content based 3D model retrieval
(CB3DMR). Current studies concentrate on the relevance
feedback or supervised learning to reduce the semantic gap
between 3D model low-level features and high-level semantic. In
this paper, a new method in extracting semantic feature for 3D
model is proposed. It can get high-level semantic information
automatically from low-level. First, invariant descriptors are
extracted from 3D models to efficient semantic annotation. An
unsupervised learning method to describe the semantics of the 3D
models is proposed. Second, and based on the resulting semantic
annotation, 3D models are semantically classified. Finally, spatial
relationships are used to derive other high-level semantic features
to augment our knowledge base and to improve the retrieval
accuracy. An ontology based 3D model retrieval framework is
used to combine the semantic concepts and visual features and
introduce the query by semantic example.