A di: -), a
ts dy, wl, wi, wy
1/2
-1/8
-AJ3
-8/36
1 coefficients.
al differentiation
f labor and time
e we do not need
e number of par-
n shows that the
nts to less than
Or.
tor: The least-
um of the square
in the sense that
ue is 0) and effi-
he Cramér-Rao's
sume below that
pendent 0 mean
erms are negligi-
ularity.
1 (8) and the un-
im likelihood es-
Vilks, 1962], and
ent in the linear
y, it is enough to
od estimator.
collinearity con-
(; — Ko) Q
e processing (the
tropic; etc), the
it "c2. P 4 72. Q?
ndependent of :.
x in (2) as we did
Ters from the log-
a constant. This
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 5. Hakodate 1998
APPLICATION OF RESOLUTION ENHANCEMENT ALGORITHM TO CLOSE-RANGE SURFACE MODELLING
Kerry McIntosh
John Fryer
Department of Civil, Surveying and Environmental Engineering
University of Newcastle
Australia
Commission V, Working Group 1
KEY WORDS: Resolution enhancement, algorithm, image matching, surface modelling.
ABSTRACT
The enhancement of digital images is a research topic which is being reported with increasing frequency in image and signal
processing journals. The applications in which these algorithms are being used include medical imaging, industrial applications
and for the conversion of video images to high-definition television (HDTV) format.
The purpose of this paper is to present the results of a practical close-range application of an algorithm used to enhance the
resolution of digital images. Several images of low resolution were combined to form a higher resolution image. The technique
involved using a combination of area-based matching to register the images with respect to one another and a rigorous
mathematical model to form the new image. Digital terrain models were generated over a test object using both raw digital
images and those enhanced by the algorithm. The resulting surface models were compared for three dimensional accuracy. The
experimentation had to take many factors into consideration, including control points, illumination and texture of the object
surface to optimise the results from the image matching.
Further discussion is presented on the future research directions for refining this algorithm, including alternative matching
algorithms and the incorporation of additional parameters into the enhancement model.
1. INTRODUCTION
Digital photogrammetry can be a fast, efficient and cost
effective method of obtaining three dimensional information,
depending on the requirements of the application to which it is
being applied. The automation in digital photogrammetry
reduces the amount of manual labour required from the
traditional stereo-plotter operator. Digital photogrammetry
will not fully replace traditional photogrammetry due to the
many limitations which have not yet been overcome. More
likely, the two technologies will be integrated to take
advantage of the best points of each. This integration will
continue until digital methods can provide sufficient accuracy
at a reasonable cost.
The limitations to digital photogrammetry include the need to
store large amounts of digital data, insufficient or overly
expensive computing power, the expense of acquiring images
at sufficient resolution, and insufficient accuracy for the
requirements of the application. Many of these hurdles have
been overcome with modern technology, but still require
further improvements before traditional photogrammetric
methods will be completely superseded (Saleh, 1996).
Imagery for digital photogrammetric applications can be
acquired in several ways, including from digital still cameras
and analog video cameras in combination with a frame
grabber. The accuracy achievable in an application is related
to the resolution of the digital imagery used. The lower the
resolution of the imagery, the lower the level of accuracy
attainable. Depending on the requirements of the application,
the resolution also affects the visual quality of the results and
the precision of classifications made from the imagery. This
limitation has been noted by several researchers, including
117
Motala, 1997; Uffenkamp, 1993; Wong and Obaidat, 1994,
and is the subject of investigation by the authors of this paper.
Scanning images from film is an alternative method of digital
image acquisition, with unique considerations and limitations,
such as film distortions, scanner distortions and processing
time from film to digital form. Scanning is regularly used in
applications such as mapping from aerial photography.
Despite many improvements in scanning technology, the
quality of scanned images does not yet equal the quality of
aerial photographs (KólIbl and Bach, 1996).
Many applications, particularly close-range, require the speed
and on-line capabilities of analog video cameras, or the
portability and flexibility of digital still cameras. However,
digital cameras with high resolution sensors are expensive and
inaccessible to many users. Often, a lower resolution camera
is used rather than one which would produce the best results
for the application. Thus the accuracy achievable is also
reduced.
1.1 Scope of Research
After noting the limitations of digital photogrammetry which
are imposed by insufficient resolution in the imagery, it was
concluded that an inexpensive method of attaining high
resolution imagery was required. An algorithm was developed
to use several low resolution images captured by an
inexpensive digital camera. These images were combined to
produce a higher resolution image. The algorithm was based
on a model of a static scene with a dynamic or portable
camera, thus allowing small shifts between each of the low
resolution images. This provides the basis of the solution of
the algorithm presented.
