<XML><RECORDS><RECORD><REFERENCE_TYPE>3</REFERENCE_TYPE><REFNUM>8678</REFNUM><AUTHORS><AUTHOR>Lo,T.W.R.</AUTHOR><AUTHOR>Siebert,J.P.</AUTHOR><AUTHOR>Ayoub,A.F.</AUTHOR></AUTHORS><YEAR>2007</YEAR><TITLE>An Implementation of the Scale Invariant Feature Transform in the 2.5D Domain</TITLE><PLACE_PUBLISHED>10th International Conference on Medical Image Computing and Computer Assisted Intervention, Brisbane, Australia, Workshop on Content-based Image Retrieval for Biomedical Image Archives: Achievements, Problems, and Prospects, October 29th, 2007.</PLACE_PUBLISHED><PUBLISHER>N/A</PUBLISHER><LABEL>Lo:2007:8678</LABEL><KEYWORDS><KEYWORD>SIFT</KEYWORD></KEYWORDS<ABSTRACT>This paper presents a complete algorithm that extends the two-dimension (2D) Scale Invariant Feature Transform (SIFT) to the 2.5 dimensional (2.5D) domain. Robust feature descriptors are extracted from the range images of human faces around stable keypoint locations, detected within a multi-scale signal representation of a [0,1] z-normalised range image. The extracted feature descriptors encapsulate not only the local image gradient orientations within each sampling window at an appropriate scale, but also the relative frequencies of the surface types present, derived using the bounded [0,1] shape index. Nine overlapping Gaussian weighted (=1.0) sub-regions are placed within each measure- ment sampling aperture in order to minimise aliasing effects caused by spatial sampling. As a result, the feature descriptors extracted are stable and exhibit good invariance to out-of-plane viewpoint rotational changes.</ABSTRACT><NOTES>accepted</NOTES><URL>http://www.miccai2007.org/index.html</URL></RECORD></RECORDS></XML>