一种鲁棒的基于光度立体视觉的表面重建方法

提出一种基于先进的凸优化技术的光度立体视觉重建框架. 首先通过鲁棒的主成分分析(Robust principle component analysis, RPCA)祛除图像噪声, 得到低秩矩阵和物体表面向量场, 然后再通过表面重建算法从向量场来恢复物体形状. 相对于先前的一些使用最小二乘或者一些启发式鲁棒技术的方法, 该方法使用了所有可用的信息, 可以同时修复数据中的丢失和噪声数据, 显示出了较高的计算效率以及对于大的稀疏噪声的鲁棒性. 实验结果表明, 本文提出的框架大大提高了在噪声存在情况下物体表面的重建精度.

A Robust Approach Based on Photometric Stereo for Surface Reconstruction

We present a new framework for surface reconstruction with technique of photometric stereo, which is based on advanced convex optimization technique. We firstly remove the errors in images by robust principle component analysis (RPCA), and then obtain low-rank matrix and surface normal field. Unlike previous approaches, this method uses all the available information to simultaneously fix missing and erroneous entries. The new technique is more computationally efficient and provides theoretical assurance for robustness to large errors. Experimental results demonstrate that this framework can improve the precision for surface reconstruction with noise.