具有向量場控制功能的實體材質合成

Abstract

目前對於實體材質合成(solid texture synthesis)的研究，大多以二維切面(2D slice)來模擬三維合成(3D synthesis)的效果，然而此類方法只記錄立體空間中，三個二維切面(2D slice)上的資料，用做鄰近點比對(neighborhood matching)，對於在立體空間中的其他部份資料則無法拿來利用於材質合成(texture synthesis)的過程中，使得最後結果會喪失空間中的資料，更甚至無法控制立體空間的材質(texture)方向性。本篇利用立體空間中的鄰近點比對(neighborhood matching)，發展出一套可以運用在真正三維空間中實體材質合成(solid texture synthesis)的方法；過程中使用表面向量值(appearance vector)取代傳統只用色彩值(RGB color value)來做鄰近點比對(neighborhood matching)，有了資料量豐富的表面向量值(appearance vector)，我們就可以只比對8個點來建立5×5×5個點的立方體(cube)內的鄰近點(neighborhood)資料，並且利用額外的向量場(vector field)來達到控制材質方向性(texture control)的目的。

Recently, some researches have been focusing on solid texture synthesis and most of them use three orthogonal 2D slices to synthesize solid textures. However, these methods only use the information on three 2D slices for neighborhood matching, and the information within 3D space is not used. It makes the results lost some information, and it is unable to control solid textures in the 3D space. Our method presents a new technique for generating solid textures with cube neighborhood matching. It helps us to synthesize within real 3D space. Appearance vectors are used to replace color neighborhood values. With these information-rich vectors, we can only use 8 locations acts for 5×5×5 cube structure neighborhoods. Additionally, we introduce our approach for controllable texture synthesis with vector fields for coherent anisometric synthesis.