Catalogue of Artificial Intelligence Techniques
Keywords: Gaussian curvature, principal curvature
Categories: Pattern Recognition and Image Processing
Author(s): Robert B. Fisher
A form of Region Finding that organises data from a range image into patches with a coherent property, usually based on some form of local surface shape. The two principal curvatures principal curvature or the mean and Gaussian curvatures are often used for purely data-driven segmentation; that is, all adjacent image pixels with similar curvatures are grouped to form a patch. Refinements include using only the signs of the curvatures to produce larger patches and to overcome noise, and prohibiting merging across depth discontinuities. Another refinement fits quadratic (or higher order) surfaces to stable patches and then incrementally grows the regions by adding image pixels that lie close to the estimated surface.
- Besl, P., Surfaces in Range Image Understanding
, Springer-Verlag, New York and London, 1988.