Automatic upright orientation and good view recognition for 3D man-made models
Pattern Recognition
| Hi-index | 0.00 |
This paper proposed a novel method for selecting good views in 3D scene Observation. A lot of existing methods for computing viewpoint quality are based on the number of visible faces and the projected area only, so they miss the visual identity of the scene. Our method combines the notion of mesh saliency and fuzzy nearness. First, we construct a fuzzy set for a viewpoint. The elements of the set are faces in the scene, and membership function is based on the visibility of faces. Then, we define an ideal fuzzy set, and the membership degree is in proportion to the mesh saliency weighted actual area of the face. Finally, we take the fuzzy nearness of these two fuzzy sets as the measure of viewpoint quality. In addition, we design a CUDA method, which avoids data transfer from device memory to host memory. So this method improves the efficiency in the computing of the viewpoint quality. We give some experiment results, and these results show that our method works well.