| Texture classification is a very important branch in pattern recognition field, and its purpose is to distinguish the difference of all kinds of objects. It is widely used in computer vision, image processing and computer graph. Work related with 3D surface texture such as data collection, analysis and classification become an active research topic in the field. Now technology of 3D surface texture classification is an advanced orientation in further research.Generally speaking, some prior knowledge should be acquired before classification, and then a model should be established according to the prior knowledge. Furthermore, certain effective methods need to be used to increase classification rate. There are three main parts in this work: model establishment, feature parameter extraction and selection, and classifier design.In this thesis, several methods about feature selection and classification in this field are first introduced, including some background knowledge, and the 3D surface texture classification and directionality research is further discussed. Due to the fact that feature extracted based on wavelet-packet decomposition can effectively detect characteristics of different textures, two algorithms are discussed to accomplish 3D surface texture classification: SVM and Adaboost. Next, a new method called Active Basis is employed to extract illumination direction feature, for basis generated from the model can show directions of an image exactly. Then an algorithm about illumination estimation is proposed based on each base's histogram. According to the experimental result, the proposed method can successfully classifies multiple different textures and distinguish the illumination directions. It is non-trivial to develop an application about 3D surface texture based on this the framework. Future work will focus on inventing new parameters to increase classification accuracy and more widely served for various kinds of textures. |
PDF Full Text Request