从单幅图像获取三维场景材质风格

3D Material Style Extraction from a Single Image

材质设计是三维场景建模的重要内容,不仅耗时而且费力.为了从图像中获取三维场景的材质风格,提出一种自动从单幅图像获取三维场景材质风格的算法.首先对三维场景采样多个视图,利用反绘制技术从样本图像中自动获取材质风格,并将其转换到三维场景的各个视图中作为场景中物体的候选材质;然后定义候选材质绘制结果与样本图像材质分布差异的目标函数,其中包含两者色度的均值、方差以及直方图分布的差;最后利用禁忌搜索方法优化该目标函数,从候选材质中选取最合适的一组作为最终结果.通过用户调查并与人工交互的结果进行对比,验证了文中算法的可行性.

Material design is an important component of 3D scene modeling, which however costs a lot of time and labors. In this paper, we present a new method for automatically extracting the material style from a sample image for 3D scenes. Given a 3D scene and a sample image, we first obtain some candidate materials using inverse rendering techniques under a set of pre-determined views. Then, we define an objective function to estimate the difference of the material style between the sample image and the synthesized images rendered with candidate materials. This objective function takes into account the difference of the mean, the variance and the histogram of the chromatic distribution. Finally, we select best materials as candidate materials by optimizing the objective function using the Tabu Search method. To validate the effectiveness of our method, we compare our results with those obtained with user assistances to demonstrate our method.