Optical splitting trees
for high-precision monocular imaging


This paper introduces a framework to design optical splitting trees to perform computational photography tasks that require many sensors with a common optical axis. As the number of sensors increases, designing a good optical system is often difficult because the components have non-ideal characteristics. We describe an optimization tool takes this into account and finds ``good'' designs as specified by several weighted coefficients. Assisted by this optimizer, we demonstrate high-dynamic range, focusing, matting, high-speed, and hybrid imaging implemented on a single, reconfigurable camera containing eight sensors.


Selected images from the paper:

Our generic eight-view splitting tree system

(top) Economical band pass and (bottom) efficient dichroic mirror multispectral trees.

Actual efficiency curves for nominally 'uniform' components

'Right rotation' transformation

HDR video trees

Frames from two HDR sequences

240 fps video of a soda can opening

Hybrid high speed, multimodal visible + IR camera

Selected images from the project:

Mobile optical laboratory

120fps video frames


 author = {Morgan Mc{G}uire and Wojciech Matusik and Hanspeter Pfister and Billy Chen and John F. Hughes and Shree K. Nayar},
 title = {Optical Splitting Trees for High-Precision Monocular Imaging},
 journal = {IEEE Comput. Graph. Appl.},
 volume = {27},
 number = {2},
 year = {2007},
 issn = {0272-1716},
 pages = {32--42},
 doi = {http://dx.doi.org/10.1109/MCG.2007.45},
 publisher = {IEEE Computer Society Press},
 address = {Los Alamitos, CA, USA},