Optical splitting trees for high-precision monocular imaging


Morgan McGuire Wojciech Matusik Billy Chen John F. Hughes Hanspeter Pfister Shree Nayar
Williams College MERL Stanford University Brown University MERL Columbia University

To Appear in
IEEE Computer Graphics and Applications, Special Issue on Computational Photography


Top and side images of our eight-view splitting tree system. The superimposed beam shows the optical path.




Beam splitting is widely used to create multiple geometrically similar but radiometrically controlled views of a scene. However, acquiring a large number of such views is known to be a hard problem. We introduce the notion of an optical splitting tree that can recursively split a monocular view of a scene a large number of times. In this tree, the internal nodes are optical elements like beam splitters and filters, and the leaves are video sensors. Varying the optical elements allows us to capture at each virtual pixel multiple samples that vary not only in wavelength but also in other sampling parameters like focus, aperture, polarization, exposure, sub-pixel position, and frame time. We present a framework to design and evaluate splitting trees, a configurable hardware system that captures up to eight views, and an algorithm for automatically discovering good designs that meet a specification. The algorithm employs an optimizer that considers light efficiency, accuracy, and cost. We demonstrate several examples of both user and automatically-designed trees and demonstrate their utility for high dynamic range, multi-focus, high-speed, and hybrid high-speed/multispectral video.



Morgan McGuire, Wojciech Matusik, Billy Chen, John F. Hughes, Hanspeter Pfister, and Shree K. Nayar. Optical splitting trees for high-precision monocular imaging. IEEE Computer Graphics and Applications. March 2007.


Adobe Acrobat PDF (2.1 MB)


Divx (18 MB)