computer vision: reflectance analysis for image understanding
TRANSCRIPT
Michael Oren’s Thesis
Damian Gordon
Reflectance Analysis for Image Understanding
How do we model these surfaces ??????????
Reflectance Mechanisms
• Diffuse Reflection (Body Reflection) Lamertian Model• Specular Reflection Specular Spike (smooth) Specular Lobe (rough)
Reflectance Components
Diffuse Reflectance
Lambertian Model
• Essentially, surface brightness is independent of viewer direction and is determined solely by enery flux of the incident light.
Surface Roughness
Surface Roughness
• Assume Surface is a collection of planar facets
=> At high magnification, each pixel images one facet
=> At low magnification, each pixel images many facets
Surface Roughness
Surface Roughness
• Modelled as a series of V-cavities whose upper edges are in the same plane.
• The width of each facet is assumed to be small compared to its length.
• Torrance-Sparrow Model (1967)
Geometric Attenuation Factor (GAF)
• If surface not illuminated from normal direction, suffers from shadowing and masking
• SHADOWING : Facet partially illuminated, adjacent facet casts shadow
• MASKING : Facet partially visible, adjacent facet occludes.
Geometric Attenuation Factor (GAF)
Interreflection Factor
• Experiments suggest too important to be ignored in model
• But, energy in incident light ray diminishes after each bounce
• So, only two-bounce interreflections modelled
Interreflection Factor
Surface Types Modelled
• Uni-directional Single-Slope Distribution• Isotropic Single-Slope Distribution• Gaussian Distribution
Gaussian Distribution
Qualitative Model
• Model is being used to calculate surface orientation, reflectance and roughness, etc.
• To be tractable, model must use maths that can be easily inverted
• This is why Lambertian is so popular after 240 years
• Must sacrifice accuracy
Qualitative Model
Experiments
Wall Plaster
Sand Paper
Wood Shaving
Conclusions
• Rough surfaces are inherently non-Lambertian in reflectance
• New model could be used in graphics for realistic rendering
• Model can be used to measure surface roughness
• If rough object can appear flat
Future Work
• Simultaneous recovery of shape and roughness
Specular Reflectance
Two Main Problems
• Detecting specularitiesReal FeaturesVirtual Features• Shape recovery of specular surfacesMany pitfalls
Result not on Surface !!!!!
Real & Virtual Features
Approach
• Use virtual features to help recover shape• Model in 2D domain first • And generalize to 3D domain
Recovery of 2D Surface Profile
• Curve Representation• Important to simplify analysis• Cartesian Coordinates - represents curve as
series of points (but no local slope)• Legendre Transform – represents curve as
an envelope of tangents
Curve Representation
Caustics
• Virtual feature travels on a surface producing a family of reflection rays
• The envelope defined by the family is called the caustic
• Use caustic compactness to distinguish real from virtual feature
Caustics
2D Surface Recovery
• Develop equations to recover profile to one parameter family of equations
• May determine correct profile by tracks two virtual features
Results
Results
3D Motion
• Generalizes 2D concepts of caustics & equations
3D Motion
3D Caustics
Results
Conclusions & Future Work
• Closed-form relationship between image trajectory of a virtual feature and surface profile
• New technique for detecting virtual featuresFuture Work• Fusion of real & virtual feaure• Shape recovery of rough surfaces