+ All Categories
Home > Documents > Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip...

Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip...

Date post: 21-Jan-2016
Category:
Upload: gwendolyn-rosanna-byrd
View: 213 times
Download: 0 times
Share this document with a friend
49
Inferring Reflectance Inferring Reflectance Functions Functions from Wavelet Noise from Wavelet Noise Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science
Transcript
Page 1: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Inferring Reflectance FunctionsInferring Reflectance Functionsfrom Wavelet Noisefrom Wavelet Noise

Pieter Peers

Philip Dutré

Pieter Peers

Philip Dutré

June 30th 2005

Department of Computer Science

Page 2: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Image-based Relighting / Environment MattingImage-based Relighting / Environment Matting

Scene(fixed viewpoint)

Page 3: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Image-based Relighting / Environment MattingImage-based Relighting / Environment Matting

Scene(fixed viewpoint)

Novel Incident Illumination

+

Page 4: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Image-based Relighting / Environment MattingImage-based Relighting / Environment Matting

… …

Scene(fixed viewpoint)

Novel Incident Illumination Compute Relit Image

+ =

Page 5: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Image-based Relighting / Environment MattingImage-based Relighting / Environment Matting

… …

Scene(fixed viewpoint)

Novel Incident Illumination Compute Relit Image

+ =

ReflectanceFunction

Page 6: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Examples of Reflectance FunctionsExamples of Reflectance Functions

Diffuse BallSpecular Ball

Page 7: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Examples of Reflectance FunctionsExamples of Reflectance Functions

Diffuse BallSpecular Ball

Page 8: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Examples of Reflectance FunctionsExamples of Reflectance Functions

Diffuse BallSpecular Ball

Reflectance Function Reflectance Function

Page 9: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Reflectance Functions (frequency domain)Reflectance Functions (frequency domain)

Diffuse BallSpecular Ball

Reflectance Function (frequency domain) Reflectance Function (frequency domain)

Page 10: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Reflectance Functions (wavelet domain)Reflectance Functions (wavelet domain)

Diffuse BallSpecular Ball

Reflectance Function (wavelet domain) Reflectance Function (wavelet domain)

Page 11: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Relight a PixelRelight a Pixel

Novel Incident IlluminationSpecular Ball

Relit pixel value?

Reflectance Function (wavelet space)

Page 12: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Relight a PixelRelight a Pixel

Novel Incident IlluminationSpecular Ball

Reflectance Function (wavelet space) Incident Illumination (wavelet space)

Page 13: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Relight a PixelRelight a Pixel

Novel Incident IlluminationSpecular Ball

Reflectance Function (wavelet space) Incident Illumination (wavelet space)

( )

Page 14: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Relight a PixelRelight a Pixel

Novel Incident IlluminationSpecular Ball

Reflectance Function (wavelet space) Incident Illumination (wavelet space)

( )

Onlynon-zero

coefficients

Page 15: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Directly Observing Reflectance FunctionsDirectly Observing Reflectance Functions

Controlled Incident IlluminationPhotograph of Specular Ball

Emit(e.g. from CRT)

Page 16: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Directly Observing Reflectance FunctionsDirectly Observing Reflectance Functions

Controlled Incident IlluminationPhotograph of Specular Ball

ReflectanceFunction

(unknown)

Observed pixel

Controlled Incident Illumination (wavelet space)

Page 17: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Directly Observing Reflectance FunctionsDirectly Observing Reflectance Functions

Controlled Incident IlluminationPhotograph of Specular Ball

Unknown Reflectance Function (wavelet space)

( )

Controlled Incident Illumination (wavelet space)

Page 18: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Directly Observing Reflectance FunctionsDirectly Observing Reflectance Functions

Controlled Incident IlluminationPhotograph of Specular Ball

Controlled Incident Illumination (wavelet space)

( )

Onlynon-zero

coefficients

Observed coefficient

Unknown Reflectance Function (wavelet space)

Page 19: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Number of ObservationsNumber of Observations

Specular Ball

Reflectance Function (wavelet space)

#Photographs=

#Illumination pixels

Incident Illumination

Page 20: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Number of Observations ProblemNumber of Observations Problem

Specular Ball

Reflectance Function (wavelet space)

Incident Illumination

1000x

1000

1000x

1000

#Photographs=

#Illumination pixels

Page 21: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Wavelet Noise IlluminationWavelet Noise Illumination

Wavelet Noise

•Normal distribution of wavelet coefficients

•Mean : 0.0

•Standard deviation : 1.0

•Rescale Wavelet Noise Pattern to fit into [0..1] range

Wavelet Noise Pattern

Wavelet Noise Pattern (wavelet space)

Advantages

•Arbitrary number of different patterns possible

•Any reflectance function gives a non-zero response

•Constant average luminance

Page 22: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Estimating Wavelet Coefficients Estimating Wavelet Coefficients

(Unknown)Reflectance Function

Wavelet Noise

Assume: positions of are knownQuestion: what are the magnitudes?

( ) =Observed

Pixel Value

Page 23: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Estimating Wavelet Coefficients Estimating Wavelet Coefficients

( ) =

Leave out zero coefficients(of the reflectance function)

Wavelet Noise (linearized)

Reflectance Function(linearized)

Observed Pixel Value

Page 24: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Estimating Wavelet Coefficients Estimating Wavelet Coefficients

= …

Multiple observations matrix-vector multiplication

Wavelet NoiseReflectance

Function

Observed PixelValues

# em

itted

pat

tern

s # observations

Page 25: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Estimating Wavelet Coefficients Estimating Wavelet Coefficients

=

Finding magnitudes : Linear Least Squares problem

… …

Wavelet NoiseReflectance

Function

Observed PixelValues

Page 26: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Estimating Wavelet Coefficients Estimating Wavelet Coefficients

=

Estimation error when onlya part is approximated?

… …

Wavelet NoiseReflectance

Function

Observed PixelValues

Page 27: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Partial EstimationPartial Estimation

+… … …= = …

Wavelet NoiseReflectance

Function

ObservedPixel Values

Page 28: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Partial EstimationPartial Estimation

According to a normal distribution

+… … …= = …

Wavelet NoiseReflectance

Function

ObservedPixel Values

Page 29: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Partial EstimationPartial Estimation

According to a normal distribution

+… … …= = …

Wavelet NoiseReflectance

Function

ObservedPixel Values

Normal Normal distributiondistribution

Page 30: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Partial EstimationPartial Estimation

+… …= = …

Wavelet NoiseReflectance

Function

ObservedPixel Values

Finding the best approximation for : Linear Least Squares problem

NoIse

Page 31: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Inferring Reflectance FunctionsInferring Reflectance Functions

Reflectance Function(2D wavelet space)

Priority Queueof Candidates

Page 32: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Inferring Reflectance FunctionsInferring Reflectance Functions

Reflectance Function(2D wavelet space)

Priority Queueof Candidates

Page 33: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Inferring Reflectance FunctionsInferring Reflectance Functions

Reflectance Function(2D wavelet space)

Priority Queueof Candidates

Page 34: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Reflectance Function(2D wavelet space)

Inferring Reflectance FunctionsInferring Reflectance Functions

Priority Queueof Candidates

Page 35: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Inferring Reflectance FunctionsInferring Reflectance Functions

Reflectance Function(2D wavelet space)

Priority Queueof Candidates

Page 36: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Inferring Reflectance FunctionsInferring Reflectance Functions

Reflectance Function(2D wavelet space)

Priority Queueof Candidates

Page 37: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Inferring Reflectance FunctionsInferring Reflectance Functions

Reflectance Function(2D wavelet space)

Priority Queueof Candidates

Page 38: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Inferring Reflectance FunctionsInferring Reflectance Functions

Reflectance Function(2D wavelet space)

Priority Queueof Candidates

Page 39: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

OverviewOverview

Record photographs

Emit

Wavelet Noise

Predetermined number of photographs

Page 40: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

OverviewOverview

Record photographs

Infer Reflectance Functions

Reflectance Function

Progressive Algorithm

For each pixel

Page 41: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

OverviewOverview

Record photographs

Infer Reflectance Functions

Compute Relit Image

Relight

Incident Illumination

Page 42: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

ResultsResults

64 Haar Wavelet Coefficients256 Photographs

Reference Photograph

Page 43: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

ResultsResults

64 Haar Wavelet Coefficients256 Photographs

Reference Photograph

Page 44: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

ResultsResults

64 Haar Wavelet Coefficients256 Photographs

Reference Photograph

Page 45: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

ResultsResults

64 Haar Wavelet Coefficients256 Photographs

Reference Photograph

Page 46: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

ResultsResults

64 Haar Wavelet Coefficients256 Photographs

Reference Photograph

Page 47: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

ResultsResults

128 Haar Wavelet Coefficients512 Photographs

Reference Photograph

Page 48: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Results: High Frequency IlluminationResults: High Frequency Illumination

Page 49: Inferring Reflectance Functions from Wavelet Noise Pieter Peers Philip Dutré Pieter Peers Philip Dutré June 30 th 2005 Department of Computer Science.

Conclusion & Future WorkConclusion & Future Work

Inferring Reflectance Functions from Wavelet Noise– No restriction on material properties– Stochastic illumination patterns– Trade-off quality versus acquisition time

Future Work– Noise filtering– Higher-order wavelets


Recommended