Finding corners

摘要

Many important image cues such as ‘T’, ‘X’ and ‘L’ junctions have a local two-dimensional (2D) structure. Conventional edge detectors are designed for one-dimensional ‘events’. Even the best edge operators cannot reliably detect these 2D features. The paper proposes a solution to the 2D problem. A mathematical proof is given to explain how the ‘corner’ detector algorithm of Harris1 estimates image curvature. Although this algorithm will isolate image L junctions, its performance cannot be predicted for T junctions and other higher-order image structures. Instead, an image representation is proposed that exploits the local differential geometrical ‘topography’ of the intensity surface. Theoretical and experimental results are presented which demonstrate how idealized instances of 2D surface features such as junctions can be characterized by the differential geometry of a simple facet model.