Sampling Analysis Cengiz Öztireli using Correlations Gurprit Singh for Monte Carlo Rendering
Point Patterns in Computer Graphics Random distributions of points with characteristics Fundamental for many applications in graphics
Imaging
Patterns of Nature
Dynamic Structures
Simulations
Geometry Processing
Fabrication
Non-photorealistic Rendering
Rendering – Computing Integrals
Estimating Integrals with Points Sample and sum the sampled values of an integrand 1 Z I := f ( x ) d x |D| D n ˆ X I := w i f ( x i ) I =1 bias P [ˆ I ] = I − E P [ˆ I ] var P [ˆ I ] = E P [ˆ I 2 ] − ( E P [ˆ I ]) 2
Stochastic Point Processes Formal characterization of point patterns
Stochastic Point Processes Formal characterization of point patterns Point Process
Stochastic Point Processes Examples of point processes Natural Process Manuel Process
General Point Processes Infinite point processes Observation window
General Point Processes Assign a random variable to each set B B B N p B q “ 3 N p B q “ 5 N p B q “ 2
General Point Processes Joint probabilities define the point process B 1 B 1 B 1 B 2 B 2 B 2 p N p B 1 q ,N p B 2 q
Point Process Statistics Correlations as probabilities % ( n ) ( x 1 , · · · , x n ) dV 1 · · · dV n = p ( x 1 , · · · , x n q Product density Small volumes Points in space dV i x i
Point Process Statistics First order product density % (1) ( x ) = � ( x ) Expected number of points around x Measures local density x
Point Process Statistics First order product density ) = � ( x )
Point Process Statistics First order product density ) = � ( x ) Constant
Point Process Statistics Second order product density % (2) ( x , y ) = % ( x , y ) Expected number of points around x & y y Measures the joint probability p ( x , y ) x
Point Process Statistics Higher order product density? z Expected number of points around x , y , z y Not necessary: second order dogma x
Point Process Statistics Higher order not necessary: second order dogma % (1) ( x ) = � ( x ) % (2) ( x , y ) = % ( x , y ) y x x
Point Process Statistics Summary: 1 st & 2 nd order correlations sufficient % (1) ( x ) = � ( x ) % (2) ( x , y ) = % ( x , y ) y x x
Point Process Statistics Example: homogenous Poisson point process a.k.a. random sampling p ( x ) = p p ( x , y ) = p ( x ) p ( y ) p ( x , y ) = % ( x , y ) dV x dV y λ ( x ) dV = p λ ( x ) = λ y = p ( x ) p ( y ) = ) = � ( x ) dV x � ( y ) dV y % ( x , y ) = � ( x ) � ( y ) = � 2
Point Process Statistics Summary: 1 st & 2 nd order correlations sufficient % (1) ( x ) = � ( x ) % (2) ( x , y ) = % ( x , y ) y x x
Stationary Point Processes Stationary Isotropic (translation invariant) (translation & rotation invariant)
Stationary Point Processes Stationary (translation invariant) λ ( x ) = λ % ( x , y ) = % ( x − y ) = λ 2 g ( x − y ) Pair Correlation Function (PCF) DoF reduced from 2d to d
Stationary Point Processes Isotropic point process (translation & rotation invariant) λ ( x ) = λ g ( x − y ) = g ( || x − y || ) g p r q PCF r
Estimating Correlations Campbell’s Theorem Z hX i f ( x i ) = R d f ( x ) λ ( x ) d x E P 2 3 Z 4X 5 = f ( x i , x j ) R d ⇥ R d f ( x , y ) % ( x , y ) d x d y E P ( x i I 6 = j
Estimating Correlations First order λ ( x ) = " X # hX i I D ( x i ) = E P 1 E P x i ∈ D # Z Z = λ d x = λ d x = λ |D| D D Point distribution P P k N k ( D ) ˆ λ = Number of point K |D| distributions
Estimating Correlations Second order stationary - pair correlation function (PCF) 2 3 4X δ ( r − ( x i − x j )) E P 5 ( x i I 6 = j Z = R d × R d � ( r − ( x − y )) % ( x − y ) d x d y Z = λ 2 R d × R d δ ( r − ( x − y )) g ( x − y ) d x d y = λ 2 g ( r )
Estimating Correlations Second order stationary - pair correlation function (PCF) 1 X X g ( r ) = ˆ δ ( r − ( x i − x j )) K λ 2 P k x i , x j 2 P k ,i 6 = j Finite domains: 1 X X g ( r ) = ˆ δ ( r − ( x i − x j )) K λ 2 a I D ( r ) P k x i , x j 2 P k ,i 6 = j
Estimating Correlations Second order stationary - pair correlation function (PCF) Point Distribution Pair Correlation Function
Estimating Correlations Second order isotropic - pair correlation function (PCF) 1 X g ( r ) = ˆ k ( r � k x i � x j k ) λ 2 r d � 1 |S d | i 6 = j Volume of the unit Kernel hypercube in d dimensions e.g. Gaussian
Pair Correlation Function g ( r ) = ˆ 1 g ( r ) = ˆ
Pair Correlation Function g ( r ) = ˆ 1 g ( r ) = ˆ g ( r ) = ˆ
Pair Correlation Function g ( r ) = ˆ 1 g ( r ) = ˆ
Spectral Statistics Power spectrum Fourier transform of PCF
Spectral Statistics Points PCF Power spectrum Points
Spectral Statistics Power spectrum Radial average Radial anisotropy
Statistics for Stationary Processes Summary Stationary: Spatial (PCF) & spectral (power spectrum) PCF Power spectrum Isotropic: radial averages
Recommend
More recommend