SVBRDF-Invariant Shape and Reflectance Estimation from a Light-Field Camera.

Area of Science:

• Computer Vision
• Computer Graphics
• Optics

Background:

• Light-field cameras offer one-shot 3D shape capture.
• Recovering shapes of glossy objects is difficult due to challenges with standard Lambertian cues like photo-consistency.
• Existing methods struggle with complex reflectance properties of materials like metals and plastics.

Purpose of the Study:

• To derive a spatially-varying (SV)BRDF-invariant theory for 3D shape and reflectance recovery from light-field cameras.
• To address the challenge of capturing glossy object shapes.
• To enable simultaneous recovery of shape and reflectance from a single light-field image.

Main Methods:

• Developed a novel analysis of diffuse plus single-lobe SVBRDFs in a light-field context.
• Derived an equation relating depths and normals, overcoming direct shape recovery limitations.
• Employed a polynomial (quadratic) shape prior to resolve depth ambiguity.
• Validated the theory using synthetic data from the MERL BRDF dataset and real-world examples.

Main Results:

• Successfully derived a theory for SVBRDF-invariant shape and reflectance recovery.
• Demonstrated that an equation relating depths and normals can be established even when direct shape recovery is not possible.
• Showcased simultaneous recovery of shape and spatially-varying bidirectional reflectance distribution functions (SVBRDFs) from single light-field images.
• Achieved accurate results on both synthetic and real-world glossy objects.

Conclusions:

• The proposed SVBRDF-invariant theory provides a robust method for 3D shape and reflectance recovery of glossy objects using light-field cameras.
• The use of a polynomial shape prior effectively resolves ambiguities in shape estimation.
• This work advances the capabilities of passive 3D shape capture for challenging materials.