A
hierarchical representation for matching deformable
P. Felzenszwalb
We consider the problem of recognizing and detecting objects
using
deformable shape models. We introduce a new representation for
two-dimensional objects which simultaneously captures geometric
information at multiple levels of resolution. The representation is based on
a hierarchical decomposition of an object's boundary. We
show how this hierarchical representation can be used in an elastic
matching framework, both for comparing pairs of objects and for
detecting objects in cluttered images. In contrast to classical deformable
boundary models our representation explicitly captures
global shape information in the form of local properties at coarse
resolutions. This leads to richer shape models and more accurate
matching results. Our experiments demonstrate state-of-the-art classification
results in several large shape datasets. We also
illustrate how the hierarchical representation can be used to detect deformable
objects in cluttered images.