CTu1C.2.pdf Imaging and Applied Optics © OSA 2013
Single Pixel Compressive Imaging of
Laboratory and Natural Light Scenes
Stephen J. Olivas,
1
Yaron Rachlin,
2,3
Lydia Gu,
2,4
Brian Gardiner,
2
Robin Dawson,
2
Juha-Pekka Laine
2
and Joseph E. Ford
1
1
Photonic Systems Integration Laboratory, Electrical Engineering Department
University of California at San Diego, 9500 Gilman Dr., La Jolla, CA 92093, USA.
2
The Charles Stark Draper Laboratory, Inc.,
555 Technology Square, Cambridge, Massachusetts 02139, USA.
3
Currently with MIT Lincoln Laboratory,
244 Wood Street, Lexington, MA 02420-9108 USA.
4
Currently with Cogito Health, 7 Water Street, Suite 400, Boston, MA 02109 USA.
Contact: Stephen J. Olivas tel: 858.822.4406 fax: 858.534.1225 email:sjolivas@ucsd.edu
Abstract: We evaluate the performance of a compressive imager using multiple basis sets on
monochromatic, color and infrared natural light scenes, and compare the results to a camera
with the same number of (conventional pixel) measurements.
© 2013 Optical Society of America
OCIS codes: 110.0110 Imaging systems, 100.0100 Image processing and 110.1758 Computational imaging.
Traditional cameras capture one measurement per pixel then discard much of the data during compression, whereas
compressive imaging (CI) systems acquire comparably fewer spatially filtered scene measurements from which to
reconstruct an image [1]. This eliminates processing unused data, making CI more efficient. Several system designs
have been considered [2] as well as custom sensor implementations [3].CI is based the idea that by sampling an image
with random basis set, the sparsest solution can be found. We explore this mode of operation as well as one in which
lower spatial frequencies are selectively sampled, making it tailored to the statistical distribution of natural images [4],
given by S( f )= f
-γ
, where 1.8 < γ < 2.3. To do so, we assembled and characterized a single-pixel CI system [5,6]
(Fig. 1) and compared it to a conventional imager using both laboratory and outdoor natural light scenes.
Fig. 1. (Left) System schematic: a scene is imaged onto a spatial light modulator (SLM) which
redirects the weighted scene energy to a photodiode for measurement acquisition. (Center) Photo of
the CI system. (Right) Photo of CI system taking outdoor natural light images.
1. System Design and Configuration
A computer running Digital Light Innovation’s Accessory Light-modulator Package (ALP) software was used to up-
load previously generated PNG transform basis image files to the Discovery D4100 controller board via a USB 2.0
link, which in turn controls the Texas Instruments digital micro-mirror device (DMD) chip. A Newport Optics Corpo-
ration photo-detector 918D-SL was used to measure light intensity collected from the DMD and send it to a Newport