| Summary: | 碩士 === 國立暨南國際大學 === 資訊管理學系 === 96 === Level-of-detail (LOD) is the one of the significant techniques which widely applied in the field of computer graphics. Simply to said, LOD is a kind of technique that a mesh of 3D model is simplified according to the distance between viewer and model in a scene. Nowadays the applications of LOD include all kinds of 3D gaming optimization, dynamic terrain rendering, etc. The purpose of LOD research is in order to alleviate the cost and improve the efficiency in computation by spitting, collapsing, decimating triangles in a mesh.
In this research, we propose a method to implement the perceptual prototype by loading several sample models in the same class. And then, this prototype is used to improve the LOD algorithm. Furthermore, a modified LOD system based on QEM algorithm is implemented.
Due to the prototype, the system can know the perceptual features and the model can reserve these perceptual features as simplifying. The perceptual concept is to recognize that which parts in the model are usually focused by human vision. If the perceptual features were destroyed in a low level of detail, it is difficult to recognize what the original model is it by human eye. Therefore, the perceptual features serve important role, while the resolution of model switching among different LODs. The system can simplify some un-perceptual parts in high level of detail. Moreover, the model can still be recognized in low level of detail before destroying perceptual parts.
The construction of the prototype is described in the following three steps. First, the branches and main body can be classified by model skeleton automatically. A model have a main body and branches more than or equal to zero. Second, loading several models and make some statistic depend on previous classification. Finally, the prototype is constructed according to the statistic data. These statistic data, such as the number of the branches and its radius and length, is analyzed by our system in a semi-automatic way. Following these steps, we can construct a skeleton-like prototype in this model class.
According to the prototype data, there are three contributions as executing LOD. First, the system can know the last level of detail as simplifying the model. The simplification of model can be recognized by human vision if the branches of model still remain in a very low resolution. Second, the system can check out the perceptual feature in a model by prototype information in a view-independent mode and reserve the perceptual parts as simplifying. Third, the system can know the recognition rate as simplification for reference.
|
|---|
