Research On 3D Model Retrieval Method Based On Shape Slicing And Feature Fusion

With the rapid development of information age and network resource sharing,3D modelshave been widely applied to the daily work area, life area, scientific research area and other areas,such as modern industry, film and television animation,3D gaming industry, biomedical industry,construction industry, archeology and so on. In order to enable users to retrieve quickly andaccurately the model in need from3D model database, researchers make thorough study on3Dmodel retrieval technology.Keyword-based retrieval requires the advance marking to all the models of the modellibrary artificially. If the description of the model is not accurate and sufficient enough, it can’tsatisfy the needs of users. The retrieval based on the content does not need artificial intervention,and it can extract its spatial structure, texture, material and other basic information automaticallyaccording to the shape features of the model, and generate a unique feature descriptor for modelretrieval. The key of the content-based3D model retrieval technology is to study the modelshape feature extraction algorithm.This paper introduces an improved3D model retrie val method based on principal planeslices and presents a new retrieval method based on multi-feature combination by summarizingthe existing3D model retrieval system and model feature extraction algorithm. This papermainly contains the following aspects:1. This paper describes the relevant content of3D model retrieval technology, whichincludes the framework constitution of the existed retrieval platforms and model pretreatmenttechnology. The existed model description feature extraction methods are s ummarized and theiradvantages and disadvantages are analyzed, including the method based on the statisticaldistribution, the method based on image, the method based on functions transformation, themethod based on topological structure, the method based on slice and the method based onmulti-feature mixture.2. This paper proposes an improved3D model retrieval method basedon orthogonal principal plane slice. To get over the disadvantages that the existing slicestechnology has a large amount of calculation and the representativeness of the chosen slice planeis not strong, this paper suggests the standard pretreatment of3D model to be queried, so that themodel has scale invariance and translation invariance. Then, multiple cuttings are made in threedirections of the model orthogonal principal plane to obtain the contour point image of the slicedplane. According to the statistical distribution features and contour area features of the slicedplane, the model’s eigenvectors can be computed. Finally,3D model retrieval can be carried outbased on the distance between the vectors. The experiment indicates that the retrieval result ofthe algorithm is reasonable with more complete and accurate rates than the existed slicealgorithms.3. This paper presents a3D model retrieval method based on multi-feature fusion. Theglobal features of the3D mode surface area distribution and the local features of the model’smean curvature are merged. First, we use ray sampling method to sample the model’s feature points. Then, we calculate the adjacent surface area of the sample points and the mean curvature,and use fourier transformation to the area distribution feature to obtain the3D model’s finialeigenvectors. Finally, we carry out model matching according to the distances between vectors toachieve3D model retrieval. The experiments show that it achieves a better retrieval effect bycombining the3D model local features and global features.