Iso-charts: Stretch-Driven Parameterization via Nonlinear Dimension Reduction Kun Zhou, John Snyder, Baining Guo, Harry Shum presented at SGP, June 2004
Goals of Mesh Parameterization Large Charts Low Distortion
Stretch-Driven Parameterization l l Advantages n measures distortion properly for texturing apps Disadvantages n requires nonlinear optimization (slow!) n provides no help in forming charts – resort to simple heuristics like planarity or compactness l Solution: apply Isomap (NDR technique) n stretch and Isomap related: both preserve lengths n eigenanalysis rather than nonlinear optimization n provides: – good initial guess for stretch optimization – good chartification heuristic via “spectral clustering”
Iso. Map [Tenenbaum et al, 2000] Data points in high dimensional space Neighborhood graph Data points in low dimensional space
Surface Spectral Analysis 2. Perform eigenanalysis on DN to get embedding coords yi
Isomap → low stretch (take first two coords) [stretch, Sander 01], L 2 = 1. 04, 222 s Iso. Map, L 2 = 1. 04, 2 s [stretch, Sander 02], L 2 = 1. 03, 39 s Iso. Map+Optimization, L 2 = 1. 03, 6 s
Isomap → good charts (spectral clustering) Analysis Clustering
Results 19 charts, L 2=1. 03, running time 98 s, 97 k faces
Results 38 charts, L 2=1. 07, running time 287 s, 150 k faces
Results 23 charts, L 2=1. 06, running time 162 s, 112 k faces
Results 11 charts, L 2=1. 01, running time 4 s, 10 k faces
Remeshing Comparison Original model [Sander 03], 79. 5 d. B Iso-chart, 82. 9 d. B