Design And Implementation Of Semantic Annotation System For 3D Models

With the extensive and increased use of 3D models, semantic-based 3D model retrieval has received growing attention in recent years. Semantic annotation connects 3D models and semantic by transforming semantic retrieval to text retrieval. It is an important method to realize semantic-based 3D model retrieval. In this thesis, the author studies the semantic-based 3D model retrieval techniques based on the semantic annotation. This work mainly includes:(1) Realization of an improved semantic automatic annotation algorithm, which takes into account simultaneously the content features correlation of models, annotation results of existing models in the model sample database and calculation rules based on WordNet semantic similarity. Experiments show that this algorithm is able to improve retrieval accuracy.(2) Proposition of a relevance feedback algorithm based on annotation. Considering the implicit mapping between content features and the semantic established by user feedback, this algorithm also considers the semantic similarity and makes a compromise between the feedbacks from both factors. Human-computer interaction elements is added by introducing feedback modifications.(3) Proposition of an integrated semantic annotation strategy on combing automatic annotation algorithm and the relevance feedback algorithm mentioned above. This strategy further improves the retrieval performance.(4) Conception of a retrieval method correspondent to the proposed semantic annotation strategy and design of a prototype of semantic-based 3D model retrieval system with human-computer interaction function. In this system, the entire semantic retrieval process are accomplished. Proposed algorithms and strategies are validated. This research sets up a good foundation for subsequent study.