摘要
当使用蒙特卡罗路径跟踪方法渲染三维场景时,由于采样不足,可能会出现噪声和细节缺失等问题.直接提升采样数水平则会产生冗余计算.为此本文提出了一种基于样本梯度的自适应渲染算法.首先在预处理阶段,获取粗渲染结果图.然后计算图像的恰可察觉差异值和熵值,进而获取控制渲染所需的阈值.随后对样本梯度进行池化和平滑滤波,然后通过对比样本差异和阈值来判断像素点是否达到收敛状态.实验结果表明,与现有的感知驱动渲染方法相比,其PSNR和SSIM分别提升了4.75%~9.98%,9.88%~17.22%.
When using Monte Carlo path tracing method to render 3D scenes,there may be issues such as noise and missing details due to insufficient sampling.Directly increasing the sampling level will produce redundant calculations.To this end,we proposed an adaptive rendering algorithm based on sample gradients.Firstly,a coarse image is rendered in the preprocessing stage.Then we calculate just noticeable difference feature and entropy of the image,and the perception threshold for rendering control.Subsequently,the convergence state of the pixels is determined by thresholding after the adaption of sample gradient.Compared with existing perception driven rendering methods,our PSNR and SSIM have been improved by 4.75%~9.98%and 9.88%~17.22%,respectively.
作者
姚伟勋
陈纯毅
申忠业
于波
胡小娟
YAO Weixun;CHEN Chunyi;SHEN Zhongye;YU bo;HU Xiaojuan(School of Computer Science and Technology,Changchun University of Science and Technology,Changchun 130022,China)
出处
《小型微型计算机系统》
北大核心
2025年第9期2176-2182,共7页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(U19A2063)资助
吉林省科技发展计划项目(20230201080GX)资助.
关键词
路径跟踪
三维场景
感知驱动渲染
恰可察觉差异值
梯度
path tracing
3D scene
perception-driven rendering
just noticeable difference feature
gradient
