https://www.na-mic.org/w/index.php?action=history&feed=atom&title=Algorithms%3ACore1Visit_May06%3AShapeAlgorithms:Core1Visit May06:Shape - Revision history2026-04-09T16:33:33ZRevision history for this page on the wikiMediaWiki 1.33.0https://www.na-mic.org/w/index.php?title=Algorithms:Core1Visit_May06:Shape&diff=3508&oldid=prevAndy: Update from Wiki2006-12-18T13:24:02Z<p>Update from Wiki</p>
<p><b>New page</b></p><div>== Martin Styner - Enhanced Correspondence and Statistics for Structural Shape Analysis ==<br />
<br />
* spherical topology (vs star-shaped)<br />
* spharm representation: can have self-intersections<br />
* template-free alignment w/ first order ellipsoid<br />
* works fine with all cortical structures<br />
* approximation error for correspondence evaluation - distance on surface<br />
** are the curvature metrics independent? if not, should not be used for optimization (overfitting the model)<br />
** soln: you can use a separate training set and a testing set (not optimal)<br />
** you can recover the surface based on C and S (you can recompute both the mean and gaussian curvature) so this doesnt really solve the dependence problem<br />
* robustness - several methods (spharm, mdl, even m-reps) give very similar results<br />
* separating training and testing datasets<br />
* bias could be introduced by the preprocessing (opening, closing, handle filling, smoothing, etc)<br />
* application to namic data - results agree with different studies<br />
* publications - Martha Shenton<br />
<br />
== Kilian Pohl - Building Shape Prior Models for Segmentation ==<br />
<br />
* diffeomorphism<br />
* two alternative ways for mapping<br />
* adding two shapes in logodd space<br />
* the black box: estimates the mean and variance<br />
* data - schizophrenia group<br />
* the size change we see can be not-growth but some other effects<br />
* comparison to Martin's work: surface vs voxel sets<br />
** implicit assumption of closest point correspondence<br />
** preservation of topology<br />
<br />
== Ross Whitaker - A Nonparametric Approach to Shape Correspondence ==<br />
<br />
* avoiding trivial solutions<br />
* avoiding introducing extra information into the population (idea in mdl)<br />
* choices about parametrization is arbitrary<br />
* particles also used in fluid dynamics literature<br />
* does initial particle configuration bias the results?<br />
** at the level detail, yes - but not the aggregation (not for the density etc, for example)<br />
* correspondence is defined by labeling particles<br />
* particles are trying to maximize entropy in a single surface<br />
* shapes are interacting to minimize the entropy in the ensemble<br />
* the need for splitting particles can be different in each shape<br />
* sensitive to alignment - can be done together w/ procrustes<br />
* can the particles reconfigure themselves?<br />
** 'triangle flipping' can happen - no guarantees<br />
** but that can be a good thing if you interpret it as a 'wrinkle'<br />
* can compare a torus and a hippocampus<br />
* neighborhood relations can be unpreserved<br />
* two free parameters: particle density and threshold for very small modes (to regularize)<br />
* local minima<br />
* mdl results not good at 3 std dev away<br />
* useful for topologically different structures<br />
* what happens if the shape has disconnected parts<br />
* so it can be used for multiple object correspondence<br />
* can be extended to volumetric instead of surface<br />
* swapping doesnt happen in 2D, but its not guaranteed it wont happen in 3D<br />
* subdivision surfaces<br />
* a shape representation vs a correspondence establishing technique<br />
* comparing particle systems with each other<br />
* the statistics are done on point data - the neighborhood info is not even used</div>Andy