HK UST * Hong Kong University of Science and Technology HK UST Modeling Hair from Multiple Views Y. Wei, E. Ofek, L. Quan and H. Shum
HK UST Hair is a defining characteristic of human appearance
HK UST Overview Related work Principle of our approach Implementation Experimental results Conclusion
HK UST Related work Manual modeling Hadap and Magnenat-Thalmann 00, Kim et al. 02 (-) Tedious user interaction
HK UST Related work Image based rendering Matusik et al. 02 (+) Photo-Realistic (-) Long capture, Large storage (-) No animation
HK UST Related work Image based modeling – Difficult due to insufficient image resolution Typically 500 pixels for 50,000 hair fibers – Hair strands, groups of hair fibers, are visible Paris et al. 04, Grabli et al. 02
HK UST Paris et al Hundreds of images Fixed view point Controlled and variable illuminations (+) Impressive results (-) Complex capturing (-) Controlled illumination (-) limited visibility
HK UST Our goals Modeling from multiple view points Easy capturing Complex hair styles Enable capturing dynamic scenes
HK UST Image Capturing Hand-Held camera Natural illumination condition
HK UST 2D image orientation of hair strands 2D orientation map Using oriented filter as in Paris’ work
HK UST 3D direction in one view 2D direction constrains the growth direction to a plane in space. Hair segment
HK UST Two images define a unique 3D direction 3D direction in two views D = N x N’ D N N’
HK UST Each image pair defines a unique direction 3D direction in multi-views Multiple images – Linear optimization D D minimizes Where ║ D ║ = 1 ΣjΣj (N j · D) 2 σj2σj2 2D orientation confidence NjNj
HK UST 3D Hair Growth according to 2D orientation Hair segment Hair fiber – a sequence of chained segments Initial growth direction – outward from scalp
HK UST Implementation Camera calibration Hair volume definition and fiber initialization Fiber growth and visibility determination Growth termination
HK UST Camera calibration Automatic calibration – quasi-dense approach (Lhuillier and Quan 05)
HK UST Hair volume S hair S hair, hair surface – Estimated from images
HK UST Approximating S hair using visual hull
HK UST Hair volume S scalp S scalp, scalp surface Generic head model Offset of S hair Ellipsoid
HK UST Hair initialization Generate root points on S scalp Define hair coverage of S scalp by projection on hair masks
HK UST Fiber Visibility Visible P – visibility determined by hair surface geometry P
HK UST Fiber Visibility P internal: Visibility is determined by the closest surface point. S hair P’ P
HK UST Fiber Visibility P internal: The deeper P is, the smoother the growth direction.
HK UST Growth Termination Exceeds pre-defined length Grows outside the hair bounding surface Large inconsistency between views S scalp S hair Visual hull
HK UST An Example of Hair Growth
HK UST An Example of Hair Growth
HK UST Experiments : Dark Hair
HK UST Smooth hair
HK UST
Curly hair
HK UST Conclusions Multi view approach Simple capture Open possibilities for hair motion capture. Automatic process (masks are still manual). High quality results.
HK UST Future work Capturing of dynamic hair using a set of multiple cameras Recovery of hair appearance
HK UST Previous work appear by courtesy of Sylvain Paris, Tae-Yong Kim and Wojciech Matusik. Some rendering provided by Florence Bertails, Hubert Nguyen Thanks
HK UST Previous work appear by courtesy of Sylvain Paris, Tae-Yong Kim and Wojciech Matusik. Some rendering provided by Florence Bertails, Hubert Nguyen Thanks
HK UST Thank You! HK UST
Hair fibers – too small to be captured 2D image orientation of hair strands Yet, hair strands are visible. 2D orientation map
HK UST Principle of our approach Using multiple view images – Ease of capturing – Strong geometry information 3D orientation from triangulation of 2D orientations Better estimation of head geometry Better visibility – Robustness from high data redundancy Automatic process
HK UST Using oriented filter as in Paris’ work Computation of 2D orientation of hair strands 2D orientation map (color coded)