高级检索
贺怀清, 赵煜桢, 刘浩翰. 增益控制策略下的梯度域路径追踪算法[J]. 计算机辅助设计与图形学学报, 2022, 34(2): 294-304. DOI: 10.3724/SP.J.1089.2022.18837
引用本文: 贺怀清, 赵煜桢, 刘浩翰. 增益控制策略下的梯度域路径追踪算法[J]. 计算机辅助设计与图形学学报, 2022, 34(2): 294-304. DOI: 10.3724/SP.J.1089.2022.18837
He Huaiqing, Zhao Yuzhen, Liu Haohan. Gradient-Domain Path Tracing with Gain Control Strategy[J]. Journal of Computer-Aided Design & Computer Graphics, 2022, 34(2): 294-304. DOI: 10.3724/SP.J.1089.2022.18837
Citation: He Huaiqing, Zhao Yuzhen, Liu Haohan. Gradient-Domain Path Tracing with Gain Control Strategy[J]. Journal of Computer-Aided Design & Computer Graphics, 2022, 34(2): 294-304. DOI: 10.3724/SP.J.1089.2022.18837

增益控制策略下的梯度域路径追踪算法

Gradient-Domain Path Tracing with Gain Control Strategy

  • 摘要: 针对梯度域路径追踪算法在像素平面均匀采样的策略会使渲染图像在不同频率区域的渲染质量差别较大的问题,提出增益控制策略下的梯度域路径追踪算法.引入平均梯度计算局部平均梯度值,筛选并标记出高频区域和低频区域,将低频区域的样本用于对高频区域增强采样;随着每轮渲染高低频区域的不断更新,自适应地对样本进行分配,实现像素平面采样的增益控制.一方面,在不同材质和光照的场景下,将改进算法和梯度域路径追踪算法进行实验对比,结果表明,文中算法使人眼敏感的高频区域渲染更加细腻;另一方面,使用均方误差、峰值信噪比和结构相似度3种图像质量评价指标评价渲染图像,并对比采样过程中3个指标的变化情况,结果表明,在保证样本数量相同的条件下,文中算法可以加快渲染的收敛速度,提高渲染效率.

     

    Abstract: Uniform sampling on the pixel plane makes the rendering quality of gradient domain path tracing differ greatly in different frequency regions. Aiming at this problem, gradient-domain path tracing with gain control strategy is proposed. The average gradient is introduced to calculate the average gradient value of partial areas, to filter and to mark the high-frequency areas and low-frequency ones, then to enhance the sampling of the high-frequency areas by samples in the low-frequency ones. With the continuous update of the high and low-frequency areas at the end of each round of rendering, the samples are allocated adaptively, so as to realize the gain control of the pixel plane sampling. On the one hand, the experiment compares the improved algorithm and the gradi-ent-domain path tracing algorithm under different materials and lighting scenes, the results show that the im-proved algorithm makes the rendering of the high-frequency areas sensitive to the human eye more delicate; on the other hand, the experiment introduces the mean square error, peak signal to noise ratio and structural similar-ity to evaluate the rendered image, and compare the trends of the above three indicators during the sampling process, the results show that the improved algorithm can accelerate the convergence speed of rendering and im-prove the rendering efficiency under the condition of ensuring the same number of samples.

     

/

返回文章
返回