Fast template matching using bounded partial correlation
Mach. Vision Appl., 13 (4):
213--221 (2003)DOI: http://dx.doi.org/10.1007/s00138-002-0070-5
Abstract
This paper describes a novel, fast templatematching technique, referred
to as bounded partial correlation (BPC), based on the normalised
cross-correlation (NCC) function. The technique consists in checking
at each search position a suitable elimination condition relying
on the evaluation of an upper-bound for the NCC function. The check
allows for rapidly skipping the positions that cannot provide a better
degree of match with respect to the current best-matching one. The
upper-bounding function incorporates partial information from the
actual cross-correlation function and can be calculated very efficiently
using a recursive scheme. We show also a simple improvement to the
basic BPC formulation that provides additional computational benefits
and renders the technique more robust with respect to the parameters
choice.
