Belle. Jasmine. Ariel. What did they all have in common? Aside from a freakishly small waist, each of the Disney heroines also had giant, perfect eyes with expressions ranging from wide-eyed excitement, to narrow-eyed suspicion. Enough’s enough, said computer graphics experts at Disney Research Zurich and ETH Zurich.
If they can figure out a way to enable people to “feel” 3D objects, and 3D-print interactive teddies, surely they can give Disney characters some important expressive abilities. So that is what they set out to do, with a complex setup of cameras and some willing human subjects.
In a paper describing the experiments entitled High-Quality Capture of Eyes, the team points that despite the human eye being “one of the central features of individual appearance”, and obviously one that is most expressive, “its shape has so far been mostly approximated in our community with gross simplifications”.
In their new approach — as described in the paper and presented at SIGGRAPH Asia 2014 last week — the team used multiple cameras and different lighting to capture all the intricacies of the human eyeball — the texture variations, colourings, veins and refractions that make it such a dynamic feature, changing every millisecond according to the environment.
They write: “We propose a novel capture system that is capable of accurately reconstructing all the visible parts of the eye: the white sclera, the transparent cornea and the non-rigidly deforming coloured iris.”
Each part behaves differently and reflects light differently, so the team decided they must be dealt with individually. The process also means a break away from the perfection of heroines of old.
Eyes are asymmetrical and contain imperfections, the team points out. And most importantly, every person’s eyes are totally different from another’s.
To get enough data, the team used six cameras, including 100mm macro lenses focussed on the iris. Flash lighting using different levels of brightness and coloured LEDs were chosen to get different effects reflecting off the cornea, and actors were asked to lie on the floor wearing a headset during the process. Eleven poses were used all together, and the whole thing took about 20 minutes to complete. Images were stitched together to begin to form the basis of the eye movement — for instance, the team found ten images was enough to get the information to replicate iris deformation, while 140 were needed in total. A facial scanning system designed to reconstruct skin was used to generate the eye model.
The main achievement has been to model how the iris responds to light, which is probably the most obvious change to the eye we all notice day to day. The team is now working on other changes to the sclera and cornea.
The technology could help bring characters to life in games and on film, potentially using the voice actors as models to bring more of their character into the medium.
The Zurich group is duly proud of the progress, saying: “We believe that the findings of this paper will alter our community’s current assumptions regarding human eyes, and our work has the potential to significantly impact the way that eyes will be modelled in the future.” “Generically modelled eyes may be sufficient for background characters, but it now takes significant effort to manually create realistic eyes for heroes and other leading characters,” commented coauthor on the paper, computer graphics PhD student Pascal Bérard. “Our reconstruction technique can greatly reduce the time spent and help increase the realism of the eye.”
