OISblog

Large format lenses by Rodenstock and Schneider seem to support Shannon numbers of around 300-500 megapixels. This potential was exploited in the “UBC ScanCam: an inexpensive 122-million pixel scan camera” by Wang and Heidrich SPIE proceedings 5301. See
http://people.cs.ubc.ca/~heidrich/Papers/EI.04.pdf

570 megapixels is some kind of magic number. First, Fermilab announced a 570 megapixel focal plane array, which will be placed in a telescope in Chile to map the distribution of dark matter. On an unrelated note, other than by coincidence, last Fall R. N. Clark estimated the Shannon number of the human visual system at 576 megapixels.

This image is data taken today by Scott McCain at Applied Quantum Technologies showing the reflectance spectra in LWIR as collected by a slit spectrometer (top) and a 79 element coded aperture spectrometer (bottom)

Loic Denis et al. make an interesting contribution to the compressive holography literature with their recent paper

Loïc Denis, Dirk Lorenz, Eric Thiébaut, Corinne Fournier, and Dennis Trede, “Inline hologram reconstruction with sparsity constraints,” Opt. Lett. 34, 3475-3477 (2009)
http://www.opticsinfobase.org/ol/abstract.cfm?URI=ol-34-22-3475

I’m at the OSA Topical Meeting on Computational Optical Sensing and Imaging this week, co-located with the annual meeting. My favorite results are super-resolution studies presented, for example, in a talk on single molecule imaging by W. E. Moerner. Moerner is collaborating with Raphael Pieston on the use of longitudinally coded PSF’s for 3D molecular imaging. I was most impressed by his images using photoactivated localization microscopy, which show far sub-wavelength lateral resolution.

On the topic of resolution, Sukhakar Prasad also gave an excellent talk on information theoretic limits. His talk suggests that frequency extrapolation works better for low space bandwidth product systems. This in turn suggests that reasonable super-resolution might be obtained in confocal systems. Of course, the combination of confocal systems with a coded Piestun PSF seems to enable 3D super-resolution.

Pixel costs below $0.000001 are essential to gigapixel imaging. Recent estimates of iPod sensor costs are below this critical number. See

iPod or netbook, cameras click for new image sensors

Today’s reading focused on analytic lens design, which may play a role in the future of wide field imaging. Interesting precedents include multiconjugate adaptive optics, which models wavefront distortion resulting from a 3D cascade of phase modulations, and beam forming with freeform lenses. The MCAO papers I read included Fusco, et al., on “Optimal wavefront reconstruction strategies for MCAO,” JOSA 2001 and Vogel and Yang, “Fast optimal wavefront reconstruction for MCAO using Fourier domain preconditioned conjugate gradient algorithm,” Optics Express 2006. The problem with MCAO for strongly aberrated field correction is that it assumes relatively weak phase distortion. A lens is capable of strong phase modulation. The general idea of a sequence of phase modulations for wide field is at the heart of the problem, however.

MCAO led somehow to the analytic lens design strategies of Rubinstein, Wolansky and Oliker, mathematicians at Indiana, Technion and Emory. Their work has focused primarily on laser beam shaping, although Rubinstein and Wolansky have previous work on general lens design, Jacob Rubinstein and Gershon Wolansky, “Differential relations for the imaging coefficients of asymmetric systems,” J. Opt. Soc. Am. A 20, 2365-2369 (2003) seems particularly interesting.

We have posted a gigapixel group photo of DISP at www.disp.duke.edu. This image was generated by combining 100 or so snapshots taken by a scanning system on a tripod. I propose that any image that can be taken by varying the sensor characteristics in time (e.g. by using a tunable filter for spectral imaging or by scanning the optical axis for wide field imaging) can alternatively be acquired in snapshot mode using generalized sampling and physical layer coding. Our hope, therefore, is to acquire gigapixel images in snapshot cameras.

Numerous web sites and software packages explore gigapixel images formed from a mosaic of smaller images. With multiscale design, we believe that gigapixel imaging will become ubiquitous.

An obvious question arises as to whether or not data systems can handle the coming data storm as image data rates increase.

Will it be useful for people to keep lifelogs? A picture of themselves every day for their entire life? A picture of their home and neighborhood? With ubiquitous autonomous imaging systems each person might generate a terabyte each day, perhaps 100 petabytes over a lifetime.

Begtrup et al. recently claim 1 billion year stability for 0.1 terabit/cm^2 storage. A lifetime would require a million square cm, or 100 square meters. Thats 10 Km of 1 cm wide tape. If the tape is 100 microns thick, the volume is 0.01 cubic meters. That’s a reasonable sized bread box of state of the art memory.

The data flood will probably come faster than the memory solution.