<XML><RECORDS><RECORD><REFERENCE_TYPE>3</REFERENCE_TYPE><REFNUM>8901</REFNUM><AUTHORS><AUTHOR>Mabaar,M.</AUTHOR><AUTHOR>Siebert,J.P.</AUTHOR></AUTHORS><YEAR>2008</YEAR><TITLE>Smoothing Disparity Maps Using Intensity-Edge Guided Anisotropic Diffusion</TITLE><PLACE_PUBLISHED>Medical Image Understanding and Analysis 2008, 2nd-3rd July 2008, University of Dundee, Dundee, Scotland.</PLACE_PUBLISHED><PUBLISHER>N/A</PUBLISHER><LABEL>Mabaar:2008:8901</LABEL><KEYWORDS><KEYWORD>Image processing</KEYWORD></KEYWORDS<ABSTRACT>We present a mechanism for improving the recovery of 3D cine models of human surface anatomy acquired by means of stereo-photogrammetry. The accuracy of the model surface relies on the initial quality of recovered disparity values. However, current matching algorithms often produce noisy and blurred disparities due to inherent physical limitations of the stereo-cine acquisition process. An approach to grouping disparity values post-matching, based on intensity gradients, is presented. This method uses the anisotropic diffusion regularisation technique known for its noise suppression and edge localisation characteristics. To preserve disparity map features while simultaneously suppressing disparity noise, the proposed method utilises anisotropic diffusion guided by edge information extracted from a co-registered intensity image. Experimental results based on synthetic data show that the proposed approach improves the quality of the disparity maps in terms of noise suppression and local discontinuity preservation. A significant qualitative improvement in quality of a reconstructed 3D face model is observed when applying the proposed algorithm to real data.</ABSTRACT><NOTES>to appear in</NOTES><URL>http://www2.wiau.man.ac.uk/caws/Conferences/46/</URL></RECORD></RECORDS></XML>