Application of semantic neighbor-based data dimension reduction method in 3D model retrieval

A data dimension reduction method based on semantic neighbor was developed to improve the recognition speed and retrieval rate of 3D model. The semantic neighbor graph of 3D model was constructed by using content-based feedback records in the process of 3D model retrieval, and the shortest path length between two arbitrary points was selected to replace that in real characteristic manifold space. The intrinsic representation for data was constructed by using multidimensional scaling algorithm in low-dimensional Euclidean space. The experiments on Princeton Shape Benchmark show that the proposed method can hold semantic relationship in low-dimensional embedding of manifold for 3D model data and achieves good performance in 3D model retrieval.