Efficient Computational Methods for Robust Multispectral Multiframe Superresolution

The proposed research considers computational models for multisensor, multiframe superresolution.

The underlying mathematical model is a Fredholm integral equation of the first kind. For the

superresolution problem, not only is the input data contaminated by noise, but the parameters

defining the linear operator are uncertain because of uncertainty in sensor alignment.

The noisy input makes the problem a candidate for Tikhanov regularized least squares. The uncertainty in

the operator make it a candidate for regularized total least squares. Moreover, the uncertainty in

the linear operator is structured, thus structured total least squares models are appropriate.

This project considers these least squares and total least squares models and appropriate methods for

regularizing them. The merits of this activity according to NSF criteria are given below.

Intellectual Merit

1. The deployment of space variant regularization to the recently proposed least squares

and total least squares models.

2. The application of total least squares algorithms and space variant regularization to the solution

of blind deconvolution problems using images acquired by multisensor arrays.

3. The extension of structured total least squares algorithms to red--green--blue (RGB) color imaging.

4. The application of wavelet superresolution algorithms to multisensor array acquired low resolution

degraded imaging.

Broader Impact

1. The superresolution problem has impact upon military, commercial, and health care

applications in diverse disciplines such as bioengineering,satellite imaging, and electrical

and computer engineering. Other intriguing possibilities include substituting expensive high resolution

instruments such asscanning electron microscopes by their cruder, cheaper counterparts and then applying

technical methods for increasing the resolution to that derivable with much more costly equipment.

2. Survey articles will be prepared to provided tutorial information on our computational methods.

This is in addition to publishing the project's results in appropriate refereed journals and conferences.

3. The PI and co-PI intend to recruit 4 REUs per year from Penn State's University Scholars program.

4. Penn State graduate students will be collaborators with the expectation of producing

Ph.D. theses related to the project.