用SURE引导自适应采样与采样颜色直方图重构

Adaptive Sampling Guided by SURE and Sample Color Histogram Reconstruction

针对现有采样颜色直方图重构缺少自适应采样方法的问题,提出一种基于Stein’s unbiased risk estimator(SURE)的像素均方误差计算方法.首先对图像空间粗采样;然后在GPU上分块计算SURE来评估图像每个像素的误差度量值,并用该值引导非均匀自适应采样;最后使用非局部多尺度滤波方法重构图像.实验结果表明,不论从图像噪声定量化指标还是视觉效果来看,该方法都极大地改进了图像绘制质量.

We present an approach for computing pixel-wise mean squared error based on Stein’s unbiased risk estimator (SURE), to perform adaptive sampling for histogram-based sample reconstruction in Monte Carlo path tracing. Firstly, the image space is coarsely sampled; then we compute SURE on GPU by block estimates error metric of each pixel, to guide subsequent non-uniform adaptive sampling. Finally, nonlocal multi-scale filtering is used to reconstruct image. The experimental results demonstrate substantial improvements in terms of both nu-merical error and visual quality.