中文摘要：

针对复杂光照条件下非凸表面形状恢复问题,采用bounding edge视觉外型模型,在本质上融合了立体重构与剪影重构方法。首先,把目标形状、反射属性以及光照度的估计统一描述为一个最小化模型;其次,把立体匹配过程转化为一维方差最小化问题,实现目标表面采样,即部分形状恢复,避免了二义性;最后,基于局部形状求解Ward反射模型,实现目标表面的完全恢复。同时采用阈值α使算法适用Non-Lambertian条件。在光度立体条件下进行了恢复三维形状的实验,证明该方法的有效性,且不受高光和阴影的约束。

英文摘要：

A method is presented for solving the problem of non-convex surface recovery under complex illumination, which combines shaping from silhouette with shaping from stereo essentially using visual-hull model called bounding edge. At first, a uniform minimum model is characterized for the estimation of object shape, reflectance and illumination. Then, the stereo match is transformed into a 1-D variance minimum issue in order to sample the object surface, which is the local shape recovery and avoids the ambiguity in stereo match. At last, the object surface is recovered fully by resolving the Ward reflectance model with local shape. At same time, a threshold α is used to extend the algorithm to Non-Lambertian condition. Experiment results in photometric stereo show that the presented method can recover object shape accurately and stably without the constraints of highlight and shadow.