Finding “Lines” of Symmetry on 3D Craniofacial Meshes Method Compute descriptors of each mesh vertex From these descriptors, learn areas that may be related to certain landmarks cluster learned areas for a given landmark find pairs of clusters for landmarks that appear on both sides of the center (e.g. eye corners, mouth corners) find clusters that appear in the middle evaluate goodness of such clusters choose the best ones and use to find the symmetry line 1
We learned the en areas and symmetry 2
Now we try 10 separate landmarks. 10 areas learned – Ten models – Nose: acal (nose side), prn, sn,se – Eyes: en, ex – Mouth: lils, ch, sto – Chin: slab 3
acal prn 4
se sn 5
en ex 6
lils sto 7
Ch slab 8
Cluster on selected areas based on position acalprnse snen 9
Cluster on selected areas based on position exslablils chsto 10
Using landmarks given by doctors as a standard to define D_plane For 22qDS data, we have doctor marked points. From their landmarks, computer the symmetrical plane 11
How to define “good” symmetric clusters A “good” pair of clusters should be symmetric to the plane A “good” single cluster should have the center on the plane 12
Select a “good” symmetry pair (or single area) acalprnse snen 13
Select a “good” symmetry pair (or single area) exslablils chsto 14
Procedure for New Images Select possible landmark areas (from learning) Clustering Pair clusters Classify “good symmetry” Get center and draw a plane using learned centers 15
Select possible landmark areas acalprnse snen 16
Select possible landmark areas exslablils chsto 17
Cluster and Pair clusters acalprnse snen 18
Cluster and Pair clusters exslablils chsto 19
Classify “good symmetry” acalprnse snen 20
Classify “good symmetry” exslablils chsto 21
Get all the centers for good pairs and single ones 22
Using only the “inliers” to get the symmetry plane 23
Final result: pre-op 4days female unilateral CL 24
Result: pre-op 3 wks female CLP 25
Result: pre-op 3m female CLP 26