DocumentCode
2853443
Title
Initial value diffeomorphic landmark matching and its application shape statistics
Author
Vailant, M. ; Miller, Michael I. ; Younes, Laurent ; Ceritoglu, C.
Author_Institution
Johns Hopkins Univ., Baltimore, MD, USA
fYear
2003
fDate
28 Sept.-1 Oct. 2003
Firstpage
306
Abstract
Summary form only given. In this work we study the diffeomorphic landmark matching problem central to the emerging field of computational anatomy (2002, Annual Review of Biomedical Engineering) for the study of finite dimensional face orbits. We present a new approach based on recent derivation of a conservation of momentum law for the geodesics of diffeomorphic flow. Once a template is fixed, the space of initial momentum - and Nd finite dimensional space for N d-dimensional landmarks - becomes an appropriate space for studying shape via geodesic flow since the flow at any point along the geodesic is completely determined by the momentum at the origin. We solve for the extremal paths (geodesics) of the kinetic energy by computing its variation with respect to the initial momentum and by applying a gradient descent scheme. Statistical analysis of a landmarked database of 100 faces is presented and we suggest a finite dimensional diffeomorphic active shape model.
Keywords
differential geometry; face recognition; image matching; statistical analysis; Nd finite dimensional space; Nd-dimensional landmarks; computational anatomy; diffeomorphic flow; finite dimensional diffeomorphic active shape model; finite dimensional face orbits; geodesics; gradient descent scheme; initial value diffeomorphic landmark matching; kinetic energy; landmarked database; momentum law; shape statistics; statistical analysis; Anatomy; Biomedical computing; Biomedical engineering; Geophysics computing; Kinetic energy; Neodymium; Orbits; Shape; Statistical analysis; Statistics;
fLanguage
English
Publisher
ieee
Conference_Titel
Statistical Signal Processing, 2003 IEEE Workshop on
Print_ISBN
0-7803-7997-7
Type
conf
DOI
10.1109/SSP.2003.1289405
Filename
1289405
Link To Document