用于整体光照的最小方差抽样  

Minimum Variance Sampling for Global Illumination

作　　者：耿中元[1] 徐庆[1] 孙济洲[1] 

机构地区：[1]天津大学计算机科学与技术学院,天津300072

出　　处：《天津大学学报》2010年第6期473-478,共6页Journal of Tianjin University(Science and Technology)

基　　金：国家自然科学基金资助项目(60879003);天津市基础研究计划资助项目(06YFJMJC00400)

摘　　要：蒙特卡罗随机抽样算法合成的图像往往带有噪声,为了减少图像的噪声,提出了一种新的图像平面自适应抽样方法.直接光照抽样和间接光照抽样的方差及耗时有很大的差异,有必要自适应分配这两种抽样的数目.每个像素每种光照的抽样数目由直接光照和间接光照的方差和耗时来决定,根据拉格朗日乘子法将直接和间接光照的抽样自适应地分配到图像平面中,使得在给定的时间里图像的方差接近最小值.根据实验结果,与已有方法相比,新方法渲染的图像的方差降低了11.5%～56.9%.Images generated with Monte Carlo random sampling are often noisy. A novel image plane adaptive sampling method has been proposed to reduce the noise of the generated image. The variances and the time consumed in direct and indirect lighting sampling are so different that it is necessary to distribute direct and indirect lighting samples adaptively. The sampling numbers for each pixel in direct and indirect lighting are determined according to variances and time consumed in the two samplings. Lagrange multiplier method is used to distribute direct and indirect lighting samples adaptively in the image plane so as to minimize the variance within the given time. Experimental results indicate that, compared with the exist- ing algorithms, variance of the image generated with new algorithm is reduced by 11.5%-56.9%.

关 键 词：蒙特卡罗 整体光照 自适应抽样 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]