Statistical Mesh Shape Analysis with Nonlandmark Nonrigid Registration

Abstract

References and Acknowledgments ['#']-] 'UTTO* T)N #UXTO* #)N '7==O*w P)W wense surface point distribution models of the human face) Proceedings of +qqq Workshop on ==#+7 .x]]-AN -OHR-X] [=S-]] =YRO*q*9O 7)N SO*; X)W Point set registrationW coherent point drift) +qqq transactions on pattern analysis and machine intelligence HxN -x .wec) x]-]AN xxXx–:O) This research has been supported by ;7U9 as project g-HIIx-H) We thank =iriama =ydlová and qva 'offmannová for allowing us the use of their facial and palatal scan datasets) Future Work =odeling highRfrequency content such as wrinkles and folds) qxplore the options of eliminating the need for landmarks altogether) Rigid Registration 9eep only shape variables by normalizing individual translationN rotation and size5 however store specimen size for use in prediction phase) Perform generalized Procrustes analysis .;P7A on landmarks) Use the transforms from ;P7 to align the meshes) Shape Prediction and Hole Filling P87 is performed on the vertex locations that were not removed) 7 model is fitted to the P8 scores and explanatory variables with linear or quadratic multivariate regression)) Prediction will not handle excluded verticesN they must be approximated separately) wisplacements of vertices on the boundaries of excluded regions are used to fit a thin plate spline .TPSA) qxcluded vertices of mean mesh are transformed using that TPS) Without zilling 'oles zilled 'oles zilled and =arked Restore specimens0 size if a study of form .shape(sizeA is desired) The Processing Pipeline Registration and correspondence search =atch rejection =odel fitting Shape prediction 'ole filling P87 +ndependent variable Predicted specimen =eshes and 4andmarks Addressed Issues +n many studiesN landmarks cannot be placed evenly or densely enough to properly fit a deformation in feature matching) +nconsistent manual mesh trimming and noisy regions cause unwanted variability) We need a processing pipeline similar to ['#']-] robust to these issues)

Topics

    0 Figures and Tables

      Download Full PDF Version (Non-Commercial Use)