| Following the Digital audio, image and video, Digital geometry shapes was recognized as the new generation of digital media, it has been widely used in many fields. Along with the rapid development of 3D acquisition and modeling technology, the number of 3D shapes that can be shared get rapid growth by geometric progression. How to organize the resources reasonably has become one of the most important research topics in the field of digital geometric shape analysis.In this paper, we use the classification of the shapes as the starting point, on the basis of semi-supervised and nonnegative matrix factorization (NMF) framework, deeply studied the two key technologies of classifying by topic or style, the related application research is also carried out. The main results of this paper are embodied in the following three aspects:(1) User-Driven 3D Shapes Dynamic Classification Based on Interactive Nonnegative Matrix Factorization, we simply use non-negative matrix factorization method to initially implement the heterogeneous shapes classified by topic; then we use the improved t-distributed Stochastic Neighbor Embedding (t-SNE) visualization method to present a clear display of the classification results by combing the characteristics of 3D shapes; in order to better reflect the user intent in the classification process, we design and develop a visualization analysis system with the ability of various user interaction, which allows to guide the computation of NMF according to user interaction on the basis of the classified results, leading to the intuitive dynamic changes for the classification results of shapes. The experimental results demonstrate that our method could achieve real-time dynamic display of classification results according to the user intent under the premise of clear visualization of large-scale heterogeneous shapes.(2) Semi-Supervised Projective Style Analysis of 3D Shapes. This paper first propose a method based on the idea of semi-supervised and using the 2D projection information of 3D shapes to complete the style analysis of 3D shapes sets, the input shapes could be classified by style, also this method is on the basis of semi-supervised and NMF framework. In addition, in the semi-supervised process, this method that accept two kinds of constraints in the way by labels or in pairwise could finish the work based on the needs of users. And this method can also localizing some of the most representative style area on each shape in no need of manual annotation. The experimental results show that the method proposed in this paper could classify various types of data sets by style.(3) Application based on style analysis. Based on the style classification, we have implemented a series of related application research. Mainly include the application of Furniture recommendation, Best view selection and Architectural style recognition. The experimental results show that the style analysis method mentioned in this paper can well support the relevant application research, also the application of this paper has a strong applicability, convenient for practical popularization. |
