Semantic tree-based 3D model retrieval using 2D sketch queries
MetadataShow full metadata
Effectively and efficiently retrieving relevant 3D models (digital representation of objects in computer) for a 2D sketch query is important for various related appli- cations. Due to the big semantic gap existing between rough sketch representation and accurate 3D model coordinates, sketch-based 3D model retrieval (SBR) is one of the most challenging research topics in the field of 3D model retrieval. To bridge the semantic gap, a semantic tree-based SBR algorithm has been proposed in the thesis. Given a 2D sketch query and a dataset of 3D models, we first build a 3D shape network (3D ShapeNet) based on the semantic tree ontology in WordNet, which is a lexical database of concepts/synsets, by classifying the 3D models into certain class nodes of the tree, according to their semantic classification/label information (i.e. semantic concepts or names). Then, we identify the semantic attributes (i.e. semantic components) that the 2D sketch query contains by 2D sketch segmentation and labeling. Finally, by measuring the semantic relatedness between the concept set of the 2D sketch components and each class node, we compute the similarity between the 2D sketch and each class node to shortlist closest class nodes as well as the relevant 3D models for the 2D sketch query. Experimental results on ten classes of sketches and models have demonstrated promising performance in bridging the semantic gap.