Research On Visual Image-based 3D Model Retrieval And Semantic Automatic Annotation Technology

Semantic retrieval is the hot spot of 3d model retrieval domain, both feature extraction and semantic annotation are the key points of semantic retrieval, computer vision image information is the most characteristic feature, Therefore, this paper focuses on visual image-based model retrieval and semantic annotation technology. This article uses the orthogonal projection image and the perspective projection images to obtain 3d model of similarity, and applies this algorithm in the semantic automatic annotation technology. The main research work is as follows:(1) Propose 3d model retrieval algorithm which based on visual orthogonal projection image. First, obtain the similarity of projected image by computing the projected image contour edge angle and the projected image outer annulus histogram which all are based on two-dimensional orthogonal projection images, then get the similarity of three-dimensional model, and expand this idea to the three-dimensional field. The experiment results show that this approach is robust against translation, scaling and rotation transform.(2) Propose 3d model retrieval algorithm which based on 6-viewpoints and 14-viewpoints's perspective projection image. Using the method of the projected image contour edge angle and the projected image outer annulus histogram to calculate the similarity of the projected image, the experiment results present increasing the number of view point will improve the accuracy of model retrieval.(3) Propose a improved semantic automatic annotation algorithm. Combine a 3d model retrieval algorithm which based on visual orthogonal projection image with semantic similarity calculation algorithm which presented by Jiang&Conrath to automatic annotate model. This algorithm improves the annotation result.(4) System Development. Design and develop a 3d model retrieval system which own function of browse, editing and retrieval, the system implements relevant retrieval algorithm of the paper. That is simple, strong interaction and expansion characteristics. This research work is partly supported by the National High Technology Research and Development Program (863 Program. No:2008AA01Z301) of China.