Texture Synthesis Given a texture, create more CS176: Texture Synthesis All examples from Wei & Levoy CS 176 Winter 2011 CS 176 Winter 2011 1 2 Texture Synthesis Texture Synthesis Don’t expect too much And amazing successes All examples from Wei & Levoy All examples from Wei & Levoy All examples from Ashikhmin CS 176 Winter 2011 CS 176 Winter 2011 3 4 How does it work? In A Nutshell Big idea Find most similar neighborhood statistical assumptions: 1 2 3 4 5 Markov random field model 6 7 8 9 10 stationarity and ergodicity t ti it d di it 11 12 find pixels with similar neighbors high dimensional point/vector scanline order (causal neighborhood) 1 2 3 4 5 6 7 8 9 10 11 12 implementation: exhaustive search CS 176 Winter 2011 CS 176 Winter 2011 5 6
Details Neighborhood Size Choices 3 3 5 5 7 7 neighborhood size 739 s 423 s 528 s hierarchy distance measure d acceleration structures 11 11 41 41 9 9 cut search space 24350 s 1020 s 1445 s CS 176 Winter 2011 CS 176 Winter 2011 7 8 Hierarchical Approach Applications Smaller neighborhoods ok What are instances of synthesis? repair (inpainting) image editing extrapolation l user control (introducing bias) single res single res 3 levels 5 5 11 11 5 5 CS 176 Winter 2011 CS 176 Winter 2011 9 10 Resources Enhancements Papers Additional ideas Wei & Levoy store coordinates not values Fast Texture Synthesis using TSVQ better for upsampling Ashikhmin A hikh i upsample, jitter, correct l jitt t multiple passes Synthesizing Natural Textures parallel subpasses check their web pages search only shifted pixel nghbd. CS 176 Winter 2011 CS 176 Winter 2011 11 12
Enhancements Transformed Exemplar Tong et al.: k-way coherence search preprocess exemplar with nearest list Lefebvre+Hoppe: appearance space distance to features as addl. data PCA on neighborhoods CS 176 Winter 2011 CS 176 Winter 2011 13 14 Example for 75D to 3D Results CS 176 Winter 2011 CS 176 Winter 2011 15 16 Results Results CS 176 Winter 2011 CS 176 Winter 2011 17 18
Recommend
More recommend