三维离散点采样表面基于全局优化的纹理合成算法

提出了一个直接在三维离散点采样表面上进行基于全局优化的纹理合成算法．算法首先对点模型表面进行基于协方差分析的递归式聚类削分，然后通过对各初始聚类进行优化调整和区域生长，得到更为均匀的计算单元块且使相邻单元块间具有重叠区域．基于所得到的计算单元块和样本纹理，在点模型表面上建立了一个全局的纹理合成能量方程，并采用最大期卑值算法迭代优化求解出一个全局的能量最小值．在全局优化算法中，通过加入的约束条件，进而给出了点模型颜色纹理的修复算法和流引导的可控纹理合成算法．由于点模型表面离散点元的稠密性，通过对离散点元的法向基于纹理合成结果进行相应的扰动，进一步增强了表面纹理的几何粗糙感．实验结果表明，该文的算法在点采样几何表面上有效地保持了生成的纹理结构和视觉效果的连续性和平滑性，得到了令人满意的结果。

Texture Synthesis for Point-Sampled Geometry Based on Global Optimization

This paper presents a novel technique for texture synthesis on point-sampled geometry using global optimization. The point set is firstly divided into uniform clusters. Each cluster partly overlaps with its neighboring clusters. The clusters are regarded as computing unit of the point set surface. Then this paper defines a global texture energy function on the point-sampled surface based on the surface clustering and the input texture sample. The texture synthesis on the point-sampled geometry is accomplished by designing an Expectation Maximization (EM)-like algorithm to optimize the texture energy. By adding special extra terms to the energy function, the authors achieve a practical texture repairing tool for incomplete point surfaces. In addition, based on 3D flow fields, the authors implement a user controllable texture synthesis algorithm for the point set surface. To enhance the roughness appearance of the geometry, the authors reconstruct the normals of the surfels for rendering according to the texture synthesis result. The experimental results show that the method can produce smooth texture synthesis effects as well as keeping the structures in the sample texture on the point-sampled geometry.