Robust super-resolution by minimizing a Gaussian-weighted L2 error norm
conference paper
Super-resolution restoration is the problem of restoring a high-resolution scene
from multiple degraded low-resolution images under motion. Due to imaging blur and noise, this
problem is ill-posed. Additional constraints such as smoothness of the solution via regularization
is often required to obtain a stable solution. While adding a regularization term to the cost
function is a standard practice in image restoration, we propose a restoration algorithm that
does not require this extra regularization term. The robustness of the algorithm is achieved by a
Gaussian-weighted L2-norm in the data mis¯t term that does not response to intensity outliers.
With the outliers suppressed, our solution behaves similarly to a maximum-likelihood solution
in the presence of Gaussian noise. The e®ectiveness of our algorithm is demonstrated with
super-resolution restoration of real infrared image sequences under severe aliasing and intensity
outliers.
from multiple degraded low-resolution images under motion. Due to imaging blur and noise, this
problem is ill-posed. Additional constraints such as smoothness of the solution via regularization
is often required to obtain a stable solution. While adding a regularization term to the cost
function is a standard practice in image restoration, we propose a restoration algorithm that
does not require this extra regularization term. The robustness of the algorithm is achieved by a
Gaussian-weighted L2-norm in the data mis¯t term that does not response to intensity outliers.
With the outliers suppressed, our solution behaves similarly to a maximum-likelihood solution
in the presence of Gaussian noise. The e®ectiveness of our algorithm is demonstrated with
super-resolution restoration of real infrared image sequences under severe aliasing and intensity
outliers.
TNO Identifier
447001
Publisher
IOP
Article nr.
012037
Source title
4th AIP International Conference and the 1st Congress of the IPIA 2008
Place of publication
Bristopl
Files
To receive the publication files, please send an e-mail request to TNO Repository.