Fractional Gaussian Fields for Modeling and Rendering of Spatially‐Correlated Media Jie Guo 1* Yanjun Chen 1 Bingyang Hu 1 Ling‐Qi Yan 2 Yanwen Guo 1 Yuntao Liu 1 1 State Key Lab for Novel Software 2 University of California, Santa Barbara Technology, Nanjing University *guojie@nju.edu.cn
Participating Media stanschaap.com
Participating Media videoblocks.com
Participating Media
Participating Media scattering
Participating Media absorption
Participating Media Independent Scattering Approximation Classical Radiative Transport Equation
Related Work (But, a lot of works have observed correlations in particle distribution)
Related Work
Related Work [Meng et al. 2015] [Müller et al. 2016] [Bitterli et al. 2018] [Jarabo et al. 2018]
Related Work [Meng et al. 2015] [Müller et al. 2016] [Bitterli et al. 2018] [Jarabo et al. 2018]
Our method General media Heterogeneity Long‐range correlations
Spatially‐Correlated Media Uncorrelated Media Correlated Media Uncorrelated Media
Spatially‐Correlated Media Correlated Media Uncorrelated Media
Spatially‐Correlated Media 𝜏 � 𝒚 𝒛 𝜕 � 𝜏 � 𝑢 � �� � 𝑓 �� 𝒚,� � Transmittance : 𝑈 𝒚, 𝒛 � 𝑓 � � � � 𝒚��� � 𝜏 � 𝒚 � 𝜏 � 𝒚 � 𝜏 � 𝒚 Extinction field :
Spatially‐Correlated Media 𝜏 � 𝒚 𝒛 𝜕 � 𝜏 � 𝑢 � �� � 𝑓 �� 𝒚,� � Transmittance : 𝑈 𝒚, 𝒛 � 𝑓 � � � � 𝒚��� � 𝜏 � 𝒚 � 𝜏 � 𝒚 � 𝜏 � 𝒚 Extinction field : We model 𝜏 � as a Fractional Gaussian Field (FGF)
Fractional Gaussian Fields Random field A collection of random variables: 𝑌 � 𝑌 𝑢, 𝜕 , 𝑢 ∈ ℝ � , 𝜕 ∈ Ω 𝑢 𝜕
Fractional Gaussian Fields Random field A collection of random variables: 𝑌 � 𝑌 𝑢, 𝜕 , 𝑢 ∈ ℝ � , 𝜕 ∈ Ω 𝑌 𝑢 � ,· 𝑢 𝜕
Fractional Gaussian Fields Random field A collection of random variables: 𝑌 � 𝑌 𝑢, 𝜕 , 𝑢 ∈ ℝ � , 𝜕 ∈ Ω 𝑢 𝑌 ·, 𝜕 � realization 𝜕
Fractional Gaussian Fields Random field A collection of random variables: 𝑌 � 𝑌 𝑢, 𝜕 , 𝑢 ∈ ℝ � , 𝜕 ∈ Ω Gaussian (random) field A collection of Gaussian‐distributed random variables. White Noise
Fractional Gaussian Fields Then what is “ fractional ”? Let 𝐽 be the integral operator 𝐽 � 1 � 𝑦 � /2 𝐽 1 � 𝑦 𝐽 𝑦 � ? 𝐽 � 𝑦 � ? , 𝑏 � 0
Fractional Gaussian Fields Then what is “ fractional ”? Let 𝐽 be the integral operator 𝐽 � 1 � 𝑦 � /2 𝐽 1 � 𝑦 𝐽 𝑦 � ? 𝐽 � 𝑦 � ? , 𝑏 � 0
Fractional Gaussian Fields With the fractional integral operator 𝐽 � �.� [ ] White Noise
Fractional Gaussian Fields With the fractional integral operator 𝐽 � �.� [ ] White Noise
Fractional Gaussian Fields With the fractional integral operator 𝐽 � �.� [ ] White Noise
Fractional Gaussian Fields Fractional Gaussian Fields
Fractional Gaussian Fields �� 𝑒 ‐dimensional FGF: � 𝑋 : white noise fractional Laplacian 𝑡 � 0 white noise The FGF family 𝑡 � 1, 𝑒 � 1 Brownian motion 1 2 � 𝑡 � 3 fractional Brownian motion 2 , 𝑒 � 1 (fBm)
Fractional Gaussian Fields �� 𝑒 ‐dimensional FGF: � 𝑋 : white noise � Hurst parameter 𝐼 � 𝑡 � � 𝐼 Correlation
Autocovariance Function �� 𝑒 ‐dimensional FGF: � 𝑋 : white noise Autocovariance function of pink noise: 𝐷 𝐼, 𝑒 Scaling term 𝑇 � Power spectral density of white noise
Autocovariance Function �� 𝑒 ‐dimensional FGF: � 𝑋 : white noise Autocovariance function of k‐fBm:
Autocovariance Function covariance matrix (1D) Pink noise (H<0) k‐fBm (H>0)
Autocovariance Function covariance matrix (1D) Pink noise Long‐range correlation k‐fBm
2D Fractional Gaussian Fields
2D Fractional Gaussian Fields 𝐼 Correlation Aggregates
Extinction Field We model 𝜏 � as a FGF : � � � var�𝜏 � 𝒚 � �? Macro‐scale: Constant Micro‐scale: FGF Fixing 𝒚 , 𝜏 � 𝒚 is a random variable with mean 𝜏 � and its variance is controlled by the FGF
Extinction Field We model 𝜏 � as a FGF : � � � � 𝒚��� � �� � 𝒚,� The line‐averaged extinction 𝜏 � 𝒚 � � � is a � � Gaussian‐distributed random variable. � � var 𝜏 � � 1 � � 1 𝜐 𝒚, 𝑢 � � 𝜐 𝒚, 𝑢 𝑢 � � � cov 𝒚 � , 𝒚 �� d𝑢 � d𝑢 �� 𝑢 � � � autocovariance function of the FGF
Extinction Field Pink noise (H<0) � � var 𝜏 � � 1 � � 1 𝜐 𝒚, 𝑢 � � 𝜐 𝒚, 𝑢 𝑢 � � � cov 𝒚 � , 𝒚 �� d𝑢 � d𝑢 �� 𝑢 � � �
Extinction Field k‐fBm (H>0) � � var 𝜏 � � 1 � � 1 𝜐 𝒚, 𝑢 � � 𝜐 𝒚, 𝑢 𝑢 � � � cov 𝒚 � , 𝒚 �� d𝑢 � d𝑢 �� 𝑢 � � �
Extinction Field Numerical verification:
Extinction Field Numerical verification: One‐point scale‐independence Pink noise k‐fBm
Extinction Field The variance of 𝜏 � 𝒚 : White noise Pink noise K‐fBm
Transmittance Ensemble‐averaged transmittance � � 𝑓 ��� � � � 𝑓 ��� � 𝑞𝑒𝑔 𝜏 � d𝜏 � Tr 𝑢 �
Transmittance Ensemble‐averaged transmittance � � 𝑓 ��� � � � 𝑓 ��� � 𝑞𝑒𝑔 𝜏 � d𝜏 � Tr 𝑢 � characteristic function 𝜒 � � 𝐣𝑢 𝜏 � � 𝜏 � 𝜏 � ~ Γ� , � Use gamma distribution for non‐negative extinction var 𝜏 � var 𝜏 �
Transmittance Ensemble‐averaged transmittance � � 𝑓 ��� � � � 𝑓 ��� � 𝑞𝑒𝑔 𝜏 � d𝜏 � Tr 𝑢 �
Transmittance Transparent
Transmittance 𝐼 � �0.2 𝐼 � 0.5 𝐼 � 0.8
Rendering Techniques Energy‐Conserving Volumetric Rendering Equation Transport kernel:
Results
Short‐range correlations Long‐range correlations
Low density
Comparing to the GBE [Jarabo et al. 2018]
Comparing to the Bitterli model [Bitterli et al. 2018]
Uncorrelated media
Correlated media
Conclusion A mathematical tool for physically‐based modeling spatial correlations in random media. Using k‐th order fBm to generate long‐range correlations. The usage of the non‐exponential transmittance functions in an energy‐ conserving RTE framework.
Thank you! Q&A
Recommend
More recommend