视觉驱动梯度域滤波重构的自适应渲染算法  

Adaptive Rendering Algorithm for Visual Driven Gradient Domain Filter Reconstruction

作　　者：郎思祺 陈纯毅[1] 申忠业 胡小娟[1] 于海洋[1] LANG Siqi;CHEN Chunyi;SHEN Zhongye;HU Xiaojuan;YU Haiyang(School of Computer Science and Technology,Changchun University of Science and Technology,Changchun 130022,China)

机构地区：[1]长春理工大学计算机科学技术学院,长春130022

出　　处：《小型微型计算机系统》2024年第2期425-430,共6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(20190302113GX)资助;吉林省教育厅科学技术研究项目(JJKH20210844KJ)资助。

摘　　要：蒙特卡罗路径追踪渲染算法结果图像往往受噪声影响,现有去噪算法效率较低且容易丢失图像细节.因此,本文提出一种视觉驱动梯度域滤波重构的自适应渲染算法.首先在预渲染阶段获取特征图像同时通过引导滤波器对特征图像进行预滤波;然后通过图像视觉显著性划分区域进行滤波重构,在显著区域利用融合图像梯度信息的双边滤波器进行平滑去噪,非显著区域利用均值滤波器进行快速去噪;最后利用SURE(Stein′s Unbiased Risk Estimator)计算像素颜色估计量的均方误差引导自适应采样.实验结果表明,与同类算法相比,本文算法可以在更短时间内渲染出具有更优质视觉效果的图像,本文算法的结构相似性(SSIM)和峰值信噪比(PSNR)均有显著提高,运行时间平均降低8.6%以上.The result image of Monte Carlo path tracing rendering algorithm is often affected by noise,the existing denoising algorithms are inefficient and easy to lose image details.Therefore,an adaptive rendering algorithm based on visual driven gradient domain filter reconstruction is proposed.Firstly,the feature image was obtained in the pre rendering stage,and the feature image was pre filtered by the guiding filter;Then the image was filtered and reconstructed by dividing the visual saliency region In the saliency region,the bilateral filter fused with image gradient information was used for smooth denoising,and in the non saliency region,the mean filter was used for fast denoising.Finally used SURE to calculate the pixel mean square error to guide adaptive sampling.Experimental results show that,compared with similar algorithms,the proposed algorithm can render images with better visual effects in a shorter time;The structural similarity and peak signal to noise ratio of this algorithm are significantly improved and the running time is reduced by more than 8.6%.

关 键 词：蒙特卡罗 三维渲染 引导滤波器 梯度 视觉显著性 自适应采样 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]