is sought. More specifically, a bound on sub-pixel pointing precision will 
be derived based on a simple image parameter. 
Resolution and precision are different but related concepts in digital 
imagery just as they are in photography. Resolution is associated with 
recognizability, while precision is associated with locatability. 
Resolution is related to the ability to distinguish two closely spaced 
objects, while precision is related to the error in estimating the distance’ 
between two (resolvable) objects. Recognizable objects within an image are 
referred to as "detail", thus we will be exploring precision of estimates of 
the geometric position of image detail. 
The geometric precision which is obtainable from a digital image may be 
restricted by a number of factors. The pixel size, scanning pattern, 
aperture shape and size, dynamic range, and character of image detail will 
all affect the precision. Knowledge of the available precision may be 
useful in many ways: The design of control points to be included in 
digital imagery, and the selection of properties of the image scanning 
equipment may be directed by their effect on the geometric precision in the 
resultant image; image 'correlation' procedures might use a measure of the 
available precision to dynamically set parameters of the algorithm; various 
algorithms may be compared on the basis of how close they come to a known 
upper bound on precision and the trade-offs among parameters such as 
aperture size, dynamic range, and sampling interval may be evaluated. 
The contents of the image plays.an important role in determining the 
available geometric precision. To illustrate this point, consider the case 
of two long straight parallel railway tracks. If the two tracks are 
resolvable, however well, then by performing a simple fitting algorithm a 
very good estimate can be made of the gauge of the railway. In contrast, 
any estimate for the length of one railway tie (considered in isolation) 
will be much less precise. 
This dependence on image content makes it unlikely that geometric precision 
can be clearly defined independent of application. The use of standard 
targets such as bar charts has been the time honoured basis of resolution 
measurements. À similar approach may be in order for general precision 
measurements in digital imagery but, for specific applications where 
objects of known shape are to be located, a more detailed analysis may be 
necessary. 
nted in section 2 for image det 
8 inten tion 1 o establish a methodology and 
ecision Rin digital imagery may be analysed ri ously. 
The model divides a pixel in regions, each of which is refered to as a 
'locale'. The number of locales ae a pixel provides a bound on their 
size, which in turn will bound the av able geometric precision. The 
development of the formal model is pre ced with an heuristic discussion, and 
followed by some illustrative example 
A formal mathematical 
absence of noise. Th 
by which geometric pr 
model is pre il in the 
am 
ta 
framework 
or 
cg 
5 
Oo 
o m 
or 
Based on the formal model, a bound is established in section 3 for the number 
of locales, and the implications to data storage are discussed in section 4. 
In section 5 a bound is established for geometric precision with noise in the 
image. The space-optimal configuration of scanner sampling interval and 
f ts per pixel is discussed.