利用權重彩色與灰階共現機率之三維紋理合成

Abstract

本論文提出了一個利用權重彩色和灰階共現機率來做三維紋理合成(3D texture synthesis)的方法。因為由輸入的貼圖中用於鄰近點比對時所擷取的特徵會影響合成結果的品質，我們首先利用彩色和灰階共現機率這兩種方法來擷取輸入貼圖中的特徵。其中彩色的部分使用表面向量值(appearance vector)取代傳統只用色彩值(RGB color value)作為特徵向量，而灰階共現機率可隨輸入貼圖的特性來選擇不同的統計量作為特徵向量。接著透過輸入貼圖的特性改變兩種方法之間的權重值。再預先計算輸入圖片中的特徵所組成的相似集(similarity set)和三維候選點(3D-candidates)來做三維紋理合成。接著利用三維金字塔合成方法(pyramid synthesis method)來完成三維紋理的合成。

In this thesis, we introduce a 3D texture synthesis method using weighted color and grey level co-occurrence probabilities. Because the extracted features from an input 2D texture can affect the quality of the result, we use color and grey level co-occurrence probabilities to extract the features. Appearance vectors are used to replace RGB color values. And various statistics are used according to the characteristic of the input 2D texture. We achieve 3D texture synthesis by using a pre-computed similarity set and 3D-candidate set. Then we can synthesize desired 3D textures by applying the 3D pyramid synthesis method.