Local Blur Estimation and Super-Resolution
Terrance E. Boult
Department of Computer Science
Department of EECS
New York, NY 10027
Bethlehem, PA 18015
Until now, all super-resolution algorithms have pre-
sumed that the images were taken under the same illumina-
tion conditions. This paper introduces a new approach to
super-resolution—based on edge models and a local blur
estimate—which circumvents these difficulties. The paper
presents the theory and the experimental results using the
We have recently proposed a new algorithm for enhanc-
ing image resolution from an image sequence  where
we compared our super-resolution results with earlier re-
search on this subject [2, 3, 4, 5, 6, 7, 8]. We showed that
the integrating resampler  can be used to enhance image
resolution. We further showed that warping techniques can
have a strong impact on the quality of the super-resolution
imaging. However, left unaddressed in  are several im-
portant issues. Of particular interest is lighting variation.
The objective of this paper is to address techniques to deal
with these issues. The examples herein use text because
the high frequency information highlights the difference
in super-resolution algorithms and because it allows us to
ignore, for the time being, the issues of 3D edges under
different views. Clearly, matching is a critical component
if this is to be used for fusing 3D objects under different
viewing/lighting conditions. This paper, however, concen-
trates on how well we can combine them presuming good
In what follows in this paper, we first present the new
super-resolution algorithm where we compare the resulting
super-resolution images with those presented in  in Sec-
tion 2. We then review the integrating resampler, an effi-
cient method for warping using imaging-consistent restora-
tion/reconstruction algorithms, in Section 3.
rithm is well suited for today’s pipelined micro-processors.
In addition, the integrating resampler can allow for modi-
fications of the intensity values to better approximate the
warping characteristics of real sensors.
The new algo-
rithms for edge localization and local blur estimation are
This work is supported in part by NSF PYI IRI-90-57951, NSF grant
CDA-9413782, and DOD MURI program ONR N00014-95-1-0601. Sev-
eral other agencies and companies have also supported parts of this
given, respectively, in Sections 4 and 5. These two algo-
rithms are both based on the imaging-consistent restora-
tion/reconstruction algorithms described in .
work is given in Section 6.