双向纹理函数稀疏采集与重建

Sparse acquisition and reconstruction of bidirectional texture functions

针对采集材质双向纹理函数(BTF)数据过程非常耗时的问题,提出一种从材质稀疏采集数据中近似重建完整BTF数据的方法.在训练阶段,将训练数据进行聚类,通过数据分解得到能够表征每个类的基字典,由优化实验设计选出稀疏采集的角度;在采样/重建阶段,只在被选择的稀疏采样角度下采集,通过最小二乘法重建出完整的BTF数据.该方法能够分别选择光源和相机的稀疏采集位置,有效地减少了实际所需的光源和相机个数.实验结果表明,该方法可以很好地从稀疏采集的数据中重建出完整BTF数据,且精度优于已有方法.

A method to approximately reconstruct the bidirectional texture function（BTF）data of materials from their angularly sparse measurements was proposed aiming at the problem that the acquisition process of BTF data is very time consuming.In the training process,the training samples were clustered and each cluster was decomposed to obtain representation bases;the sparse sampling positions were obtained using optimal experiment design.In the acquisitions/reconstruction process,only images under the selected acquisition positions were captured.Then the cluster that each sparse image belongs to was determined;the full material data was reconstructed by solving a least squares problem.Since the acquisition positions of viewing and lighting directions can be selected respectively,the proposed method can efficiently reduce the numbers of both camera and light sources.The experiment results illustrate that the method can well reconstruct fully resolved BTFs from their sparse measurements and outperforms state-of-the-art methods in terms of accuracy.