低噪声自适应突变策略的MMLT算法  

Low-noise adaptive mutation strategy for multiplexed metropolis light transport algorithm

作　　者：贺怀清[1] 王旭 刘浩翰[1] 惠康华[1] HE Huai-qing;WANG Xu;LIU Hao-han;HUI Kang-hua(Computer Science and Technology College,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学计算机科学与技术学院,天津300300

出　　处：《计算机工程与设计》2023年第4期1189-1197,共9页Computer Engineering and Design

基　　金：国家自然科学基金-民航联合研究基金项目(U1333110);国家重点研发计划基金项目(2020YFB1600101);天津市教委科研基金项目(2020KJ024)。

摘　　要：针对多通路metropolis光照传输(MMLT)算法在光照复杂区域遍历性差、突变策略不灵活、接受概率低,导致图像含有大量噪声问题,提出一种自适应突变策略的改进算法。利用收集的样本信息,增强对场景光照的初始认知;用兼顾局部与全局样本的权重函数选择建议样本;重新设计接受概率计算方法,提高样本的接受概率;通过共享像素信息,降低噪声。实验从不同角度验证,改进算法对场景具有普适性,实现了自适应采样,提高了样本的接受概率,降低了图像噪声。Aiming at the problems of the poor ergodicity,inflexible mutation strategy and low acceptance probability of the multiplexed metropolis light transport(MMLT)algorithm in complex illumination area,which lead to a large amount of noise in the image,an improved algorithm of adaptive mutation strategy was proposed.The information about samples were collected to enhance the initial recognition of scene brightness,a weight function in which both local and global information was taken into account was proposed to select proposal samples,and the acceptance probability calculation method was redesigned to increase the acceptance probability of the sample.The noise was reduced by sharing the information between pixels.Experiments verify the algorithm from different aspects.The results show that the improved algorithm has universal applicability to the scene,realizes the adaptive sampling of samples,improves the acceptance probability of samples,and reduces the image noise.

关 键 词：全局光照 光照传输算法 马尔科夫链 蒙特卡罗方法 突变策略 接受概率 降噪 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]