Peter J. Giblin

3Machine learning papers tracked by Lacuna
Is this you?

Research Overview

Peter J. Giblin focuses on computer vision and computational geometry, specifically the mathematical representation of planar shapes and curves. His work investigates affine invariant medial axes, symmetry sets, and noise-resistant skeletonization techniques. The trajectory of his research demonstrates a progression from distance-based and conic-based symmetry definitions toward robust, area-based methods for shape analysis and skew symmetry detection.

Research Trajectory

The research trajectory of Peter J. Giblin in the late 1990s and early 2000s centers on the mathematical foundations of shape representation in computer vision, with a particular emphasis on affine invariance. A recurring challenge in computational geometry and vision is representing shapes in a way that remains consistent under affine transformations, such as those caused by changes in camera viewpoint.

In 1998, Giblin led work introducing foundational concepts for affine invariant shape descriptors. In Affine Invariant Medial Axis and Skew Symmetry, he and his co-author proposed two distinct mathematical approaches to define the symmetry set and medial axis of planar shapes. The first approach utilized affine invariant distances, defining the symmetry set as the closure of the locus of points on multiple affine normals that are affine-equidistant from corresponding points on the curve. The second approach leveraged affine bitangent conics, defining the symmetry set based on the centers of conics that maintain at least three-point contact with distinct points on the curve.

Following this foundational work, Giblin's research as a senior author shifted toward addressing the practical limitations of these geometric representations, particularly their sensitivity to noise in real-world image data. Previous approaches relying strictly on Euclidean or standard affine distances often required extensive curve denoising. To resolve this, Giblin and collaborators introduced a novel, robust framework in Noise-Resistant Affine Skeletons of Planar Curves. This 2000 paper redefined the affine skeleton using an area-based metric. Under this definition, a point belongs to the affine skeleton if it is equidistant from at least two points on the curve, where the distance is measured by the minimal areas between the curve and its corresponding chords. This area-based formulation proved significantly more robust to noise, bypassing the need for explicit denoising steps.

This area-based approach was further formalized and expanded in the 2004 journal publication Area-Based Medial Axis of Planar Curves.. Serving again as senior author, Giblin and his team extended the theoretical framework to include a dynamical interpretation of the affine medial axis. They demonstrated that the medial axis points correspond to the affine shock positions of the affine erosion of the curve. The work provided a simple computational method for these structures and demonstrated their practical utility in detecting affine skew symmetry in real images.

Current Directions

Based on the provided evidence, the most recent publication from Peter J. Giblin dates to 2004. This later work consolidates his research on area-based affine invariant medial axes and their dynamical interpretations. The focus of these most recent papers is on the robust mathematical modeling of planar curves and the practical detection of affine skew symmetry in real-world image data without the need for prior denoising.

Profile updated

See something wrong? Report a mistake