This paper firstly presents methods of resolution enhancement 
of digital imagery as reported by other researchers. There are 
an increasing number of hardware and software solutions 
becoming available to enhance image resolution. 
The resolution enhancement algorithm developed by the 
authors is described and applied to a close-range application. 
Digital photogrammetry was used to create a digital terrain 
model (DTM) of a test object using original digital images and 
images enhanced by the algorithm. The accuracy of each 
DTM was analysed and compared. Over 200,000 points from 
each DTM were used in the comparison to deduce if the 
enhancement of the resolution of the digital images had 
increased the accuracy of the determined surface models. 
Factors of the experimentation which were taken into 
consideration included control points, illumination, image 
enhancement and DTM creation. A description of future 
research directions is also presented. 
2. RESOLUTION ENHANCEMENT OF DIGITAL 
IMAGES 
The objective of image enhancement is to produce an image 
which is more suitable for an application than the original 
image (Gonzalez and Wintz, 1987). The difference between 
enhancement and restoration is that restoration is a means of 
modifying the image so it becomes how it appeared originally. 
Enhancement is trying to improve the original image to give 
better visualization (Weeks, 1996), or accuracy in 
classifications or measurements. 
The enhancement being investigated in this paper is that of 
determining higher resolution images. Hardware and software 
solutions are the two main ways this enhancement can be 
achieved. Hardware techniques include increasing the size 
and number of pixels of the sensor, decreasing the pixel size, 
and modifying a camera to move the sensor by known 
amounts. Software solutions include using one or more low 
resolution images to interpolate or solve for a higher 
resolution image when there are initially unknown amounts of 
shift between images. 
2.1 Hardware Techniques 
There have been many investigations into medium and high 
resolution cameras which are commercially available and, as 
previously mentioned, relatively expensive. Kochi (et al., 
1995) reported the development of a measurement system 
using a high resolution camera with sensor dimensions of 
4096 x 4096 pixels. The system is modular, including image 
acquisition, measurement and calibration software. Peipe 
(1995) presented an investigation on the Kodak DCS460, 
which is a digital still camera with a 3000 x 2000 pixel sensor. 
The camera was found to be suitable for off-line single sensor 
applications and the relative accuracy achievable was in the 
order of 1/180,000. 
Some hardware techniques have modified cameras to produce 
high resolution digital images. Godding and Woytowicz 
(1995) described a system of providing a digital back for a 
Rollei film camera, with a sensor resolution of 2048 x 2048 
pixels, which could achieve relative accuracies of 1/150,000. 
118 
Lenz and Lenz (1993) used a method based on the accurate 
movement of the CCD array at a sub-pixel level and described 
this method of resolution enhancement as micro-scanning. 
Also discussed was a method termed macro-scanning, which 
involved mosaicking patches of digital imagery thus 
producing a large digital image at the same resolution as the 
original images. These methods can be combined and used 
simultaneously, which occurs in the ProgRes 3000 camera 
(ibid, p59). The CCD chip is actually moved inside the 
camera left-to-right and up-and-down by increments as small 
as 3um to capture multiple images of an object. These are 
integrated to give a higher resolution image, but the cost of 
adding the electronic drive system to move the array is very 
high (approximately US$30,000). 
A relatively new invention is in the Pixera Pro camera (Reis, 
1997), which uses a magneto-optical effect found in special 
glass to achieve accurate shifts in the direction of light rays 
through the lens system, such that the scene can be positioned 
onto four different areas of the sensor. This is used to create 
an image with an apparent resolution four times greater than 
the resolution of the sensor. 
2.2 Software Solutions 
Jensen and Anastassiou (1995) presented a non-linear 
interpolation scheme for enhancing the resolution of digital 
still images by determining edges within the images to sub- 
pixel level. This method is very specific in the type of images 
it can be used to enhance. Jensen intended this method to be 
used in conjunction with other methods of image resolution 
enhancement. 
Long (et al, 1993) presented a method for generating 
enhanced resolution radar images of the earth’s surface using 
spaceborne scatterometry. Although the data being enhanced 
was different from the usual images used in photogrammetry, 
the strategy is still applicable and noteworthy. The method 
was based on an image reconstruction technique which 
utilised the spatial overlap in scatterometer measurements 
made at different times. A least squares solution was 
employed, involving singular value decomposition analysis. 
The limit in the final resolution was determined by a 
combination of the noise level in the measurements and in 
their overlap. 
A notable point raised by Long (ibid) was that noise in the 
refined images increased as the resolution was improved. 
This is a problem which has to be closely monitored and 
precisely modelled in resolution enhancement techniques to 
ensure the enhanced image has not suffered any noticeable 
degradation in accuracy. 
Wiman (1993) presented a method where a scanner was used 
to acquire several images of an aerial photograph at known 
sub-pixel translations from the first image. Equations linking 
the images were set up with the unknowns being the values of 
the pixels of a combined, higher resolution image. The grey 
values of the pixels in the high resolution image were 
determined using a pseudo-inverse method as the equation 
system was under-determined. Wiman did not pursue the 
experimentation after initial results were inconclusive as the 
system of equations he developed could not be accurately 
solved. He su 
before definite c 
The image proc 
often based o 
Stevenson (199 
which used bot 
sequence of vi 
determine the p 
noted that the q 
quality of the « 
frame, this prob 
a unique solutic 
independent obj 
A method whicl 
due to non-zero 
1997). This me 
removing acqui: 
components, be 
the removal of 
image resolutio: