Local shading analysis.

Area of Science:

• Computer Vision
• Computational Imaging
• Photometry

Background:

• Image shading analysis can infer scene properties without prior knowledge.
• Lambertian surfaces and their intensity derivatives are fundamental to image formation models.

Purpose of the Study:

• To develop methods for estimating surface orientation and illuminant direction from image shading.
• To extend these methods to natural imagery using geometric constraints.

Main Methods:

• Mathematical proof establishing equivalence between image points and Lambertian surfaces with equal principal curvatures.
• Development of algorithms to recover surface orientation, surface type, and illuminant direction.
• Application of general position and regional constraints for natural image analysis.

Main Results:

• A unique combination of image formation parameters is determined by image intensity and derivatives for surfaces with equal principal curvatures.
• Algorithms successfully estimate surface orientation and illuminant direction on both synthetic and natural images.
• Qualitative surface types (planar, cylindrical, convex, concave, saddle) can be identified.

Conclusions:

• Local image shading analysis provides robust estimates of surface orientation and illuminant direction.
• The developed algorithms offer valuable insights into scene geometry and lighting conditions.
• This approach enhances understanding of image formation and scene reconstruction.