Image Measures and Metrics by Characteristic Shapes

1
Image Measures and Metrics by Characteristic
Shapes
Tom Asaki (LANL), Mark Abramson (AFIT), Rachael
Pingel (BYU), John Dennis, Jr. (Rice)
Key Idea find low-dimensional measure vectors on
whole images that preserve relevant features, and
compute metrics on feature vectors.
Example For each image intensity zi, find the
maximal area ellipse ? such that z gt zi in ? and
on the boundary of ? subject to a maximum
allowed constraint violation h(?)?(z?zi)lth0.
Key Idea Use low-dimensional parameterized
shapes to form relevant measure vectors as a
function of image intensity.
Compose measure vectors on the ellipse
description parameters (x, y, a, b, ?) and on
meaningful derived quantities (e.g. eccentricity
e and area A).
Image A Image B
Characteristic
Characteristic Sequence A
Sequence B
Measure vectors are ellipse parameter quantities
computed as functions of image intensity.
Key Idea Characteristic shapes must be low
dimensional, capture features of interest, and
be defined over most image intensities.
circle unions p3n
ellipses p5
low-? trigonometric polynomials, p10