Title :
Correlated Active Appearance Models
Author :
Tiddeman, Bernard ; Chen, Jingying
Author_Institution :
Univ. of St. Andrews, St. Andrews
Abstract :
In this paper we describe a novel algorithm for active appearance model (AAM) alignment in which the normalized cross correlation (NCC) is maximized, rather than minimizing the sum of squared errors (SSE). We use an inverse compositional approach and derive an analytic solution rather than rely on an off-the-shelf non-linear optimization routine. We show that the algorithm performs better than traditional AAM fitting in terms of accuracy, stability and speed.
Keywords :
correlation methods; image processing; least mean squares methods; optimisation; correlated active appearance models; inverse compositional approach; nonlinear optimization routine; normalized cross correlation; sum of squared errors; Active appearance model; Algorithm design and analysis; Equations; Error correction; Internet; Linear systems; Principal component analysis; Shape; Stability; Taylor series; Active Appearance Models; Normalized Cross Correlation;
Conference_Titel :
Signal-Image Technologies and Internet-Based System, 2007. SITIS '07. Third International IEEE Conference on
Conference_Location :
Shanghai
Print_ISBN :
978-0-7695-3122-9
DOI :
10.1109/SITIS.2007.92
