Research Summary


My research interests focus on the application of mathematical techniques to describe, model, and analyze physical and biological systems, with particular emphasis on novel image reconstruction methodologies and inverse problems in imaging.

Recent Projects

Fourier transform profilometry (FTP) for biometric identification: End-to-end algorithm development for a 3-D face imaging system to be deployed at security checkpoints and uncontrolled crowd areas. From left-to-right the images show FTP data (subject illuminated by sinusoidal-amplitude laser fringes), the reconstructed 3-D surface (colormap indicates height), and a rotated surface with photograph overlay (photograph captured simultaneous with fringe data).
Phase unwrapping for FTP and DH: fast robust methods to remove 2π discontinuities from 2-D phase profiles modulo 2π. In 1-D this problem is straightforward, given sufficient sampling of the signal; in 2-D it's deceptively hard. In both FTP and 3-D DH, the surface profile is computed from the phase of a complex image and must be unwrapped to form a smooth depth map. Note the discontinuity at center in the image on the left, which has been removed in the image on the right.
Automatic fringe-aware glint-removal: data pre-processing to remove specular reflections from eyeballs, glasses frames, and wet skin. Glints cause large spikes in the 3-D reconstruction and degrade biometric ID performance. Inpainting and filtering methods computationally eliminate the glint without distorting the FTP fringes, and avoid a costly optical hardware method.
Automatic spectral region extraction for FTP: algorithm development for automatic identification of the extent of the signal modulation about the fundamental fringe frequency. Care must be taken not to extract too large an area to avoid introducing artifacts and noise into the reconstructed phase. Extracting too little reduces the scale of surface feature which can be reconstructed.
Protoype unit for demonstrations: algorithms were implemented in C++ and achieved target run-time speeds with good ID performace and unattended operation. Planned a GPGPU implementation for video-rate update of 3-D imagery and biometric ID.
Proposed digital holographic 3-D mapping of coral beds: laboratory experiment imaging real (dead) coral in a tank of saltwater. Potential to use in airborne, boat, or submarine surveys of coral reefs. Left, photograph of a piece of coral. Right, 3-D surface representation (phase; color indicates depth, 2π-ambiguity not resolved -- not "unwrapped").
Multiaperture digital holography: coherent combination of several identical DH imaging systems improves image resolution without the cost (in both SWAP and $) of achieving that resolution in a single system. The image on the left is from one aperture; few groups of bars are resolvable. The image on the right is from four co-phased apertures; many groups of bars are now resolvable. See below for some details on coherently combining apertures.
Aperture cophasing for multiaperture digital holography: to coherently combine apertures, they need to be aligned to within a fraction (1/12) of a wavelength (near-infra-red is around 1 micron). While extremely challenging using physical optics, phase corrections can easily be applied computationally. The right-hand image shows six hexagonal apertures co-phased to the center reference aperture. Only three variables are needed: piston, tip, and tilt. The left-hand plot shows iterations of Newton-Raphson optimization on isosurfaces of the sharpness metric for one pair of apertures. When the sharpness of an image formed by these apertures attains a maximum, the apertures are co-phased.
Accurate identification of atmospheric turbulence distortion in simulated digital holography: the effect of atmospheric turbulence on propagation of reflected laser light is simulated by a phase screen (left). Sharpness maximization with 10-100 variables (number of Zernike polynomials) is used to correct the resulting blur. The identified phase screen (right) approaches the projection of the simulated aberration on the Zernike basis.
Removing atmospheric turbulence distortion in long-range digital holographic imagery of small objects: fast methods for data. Results of the above development are illustrated at left. The left image is distorted by (real) atmosphere; the right image has been restored using sharpness maximization in the Zernike basis. Almost no detail is visible in the uncorrected image; separation of the bars is discernable after restoration. Note also the presence of speckle noise (blobs) as this is a single (un-averaged) image.
Correction of anisotropy in digital holography through turbulent atmosphere: restoration over isoplanatic patches and stitching. The assumption of a space-invariat kernel (point-spread-function, PSF) fails when the severity of the turbulence is large. Sharpness maximization over the whole image performs poorly since it does not account for the variations in the kernel. Once can independently apply sharpness maximization over sub-regions of the image to obtain much better results. That is, one breaks the large space-variant problem into multiple smaller space-invariant problems and then stiches the results back together. The left image shows restoration with a whole-image PSF; the right image has been restored with a single PSF optimized for the much smaller dashed box.
Proposed digital holographic surface mapping of Europa (moon of Jupiter): laboratory experiments imaging wax casts of features similar to the surface of Europa. Proposed to include a 3-D mapper on a future planned NASA mission. Left, photograph of wax cast showing flow feature. Right, 3-D surface representation (phase; color indicates depth) of a portion of the wax cast.
Polarimetric digital holography: adds information to imagery to enable identification of man-made materials (vehicles and buildings) from natural features (rocks, trees, grass). Project with two high-school student interns -- won accolades at local and regional science fairs.
Two-wavelength digital holography for 3-D imaging: developed algorithms for image formation and reconstruction of 3-D surfaces. Data acquire from both laboratory and field-demo systems.
WANDER (wavelength normalized depolarization ratio): modeling polarized laser backscatter for standoff detection of bio-aerosol terror agents such as anthrax (b. anthracis and b. globigii), plague, and viruses. In particular, WANDER can distinguish anthrax from common background aerosols such as dust, diesel smoke, and pollen. Computational modelling supported and validated the theory behind laboratory and field-test results.
Optical frequency demultiplexing for high-bandwidth heterodyne laser radar: modeled performance of a Fabry-Perot filterbank for subdividing a wide spectral band into narrower bands.
Nondestructive evaluation of spot-welds via an inverse problem in thermal imaging: recover size and shape of subsurface spot-welds in automotive sheet metal assembly from thermal videos of the sheet's surface. A transducer vibrates the sheets, generating a heat source at the weld. Simplified model of 3-D heat conduction to a 2-D model and applied regularization methods to recover the source. Traces at left show ideal source and reconstruction. Images at left show ideal source (top row), simulated data (bottom row), and reconstructed source (middle row). Ph.D. thesis project arising from internship with General Motors Corp.
Robust optimal design of an optical homogenizer operating under varying input conditions: designed a phase mask to transform stochastically varying input (left plot) to an approximate box-car output (right plot) using robust stochastic optimization methods. Masters thesis project arising from internship with 3M Corp.
Molecular beam electric resonance spectroscopy: undergraduate summer research in experimental physics. Improved data collection software, repaired hardware on the apparatus, modified modeling software to incorporate faster numerical methods.


Thomas Höft's home page.
Department of Mathematics home page.
University of St. Thomas home page.

Last updated 21 Jan 2011.