quantization noise begin to play a significant role. It 
is to this scenario that the paper is addressed. 
Digital image processing begins with the acquisition 
of a digital image. Often the digital image is derived, 
that is; it is acquired by digitizing another form of 
imagery. While the results here may be applied to 
such applications, the main interest lies in primary 
digital imagery, where the scene is imaged by a 
digital camera' generating a spatial and radiometric 
(intensity or grey level) quantized representation. 
This paper begins with a discussion of primary 
digital image acquisition issues, an appreciation of 
which is essential to the realistic analysis of 
geometric fidelity of digital images. It is noted in 
particular that the typical commercial solid state 
camera uses an RS170 image transmission standard 
to relay images from the camera to the computer and 
that this signal standard seriously jeopardizes 
geometric integrity. 
The central topic of the paper is the theory of 
"locales", which were introduced at the 1984 ISPRS 
congress in Rio de Janeiro [11]. The concept has 
recently been extended to develop an optimal 
algorithm for position estimation. Following a brief 
review of locales, the salient points of the new 
algorithm are introduced. Results of simulation 
studies are reported. Some investigations with real 
imagery are underway but results are not available at 
present. 
It is apparent that the photogrammetric community, 
more than any other field of engineering, has the 
most to gain by a thorough and rigorous analysis of 
geometric precision in digital images. Efforts and 
results in this area must come from within the 
community and all available techniques should be 
enlisted in the investigations. The theory of locales 
may prove to be very useful in this regard. 
2. ACQUISITION OF QUANTIZED 
IMAGERY 
Solid state imaging arrays, such as are used in CCD 
array cameras, provide primary data acquisition of 
high quality quanitized imagery. Calibration of CCD 
arrays has shown that array element spacing is 
regular, even by photogrammetric standards (14,3,2]). 
The precise spatial sampling is due to the regularity 
and resolution of the photolithographic process used 
in the micro-electronic industry to fabricate the 
imaging arrays. The rigid and planar construction of 
the die (the term used for the tiny piece of silicon 
containing the electronics within the "chip" package) 
further enhances the geometric integrity of array 
imagers. 
An individual sensor in an array accummulates 
electrons in a potential well formed by electrodes 
overlaying the photosensitive material. The number 
of electrons generated per photon (on the average) is 
referred to as the quantum efficiency of the device. 
The efficiencies are generally quite high, sometimes 
approaching unity so that the sensors can be used for 
"photon counting" applications. A single sensor in 
an imaging array holds about 100,000 electrons in its 
"charge bucket" [6,13]. Photodetection in CCD's 
behaves like shot noise [7,81.8] so that the variance 
in the electron count when a fixed luminous flux 
falls on the sensor is equal to the expected number of 
electrons [16,13]. An effective quantization scheme 
then, may be to set the unit intensity equal to the 
RMS deviation in the charge number. With an 
expected electron count of 100 thousand, this gives 
316 quantization levels with an RMS deviation of 
one level. There are two other main sources of noise 
within the sensor; dark current and readout noise. 
Dark current noise is generated by thermal energy. 
Cooling is performed in some cameras to reduce this 
effect but 256 levels of quantization are generally 
available at room temperature with reasonable 
lighting [6]. Readout noise levels vary with the 
method used to move the charge buckets from the 
imaging array to the amplifying electronics. Terms 
such as Charge Coupling, Charge Injection and 
Plasma Coupling refer to readout methods. A typical 
readout noise level is less than 100 electrons [8,13], 
consequently readout noise is insignificant except at 
very low light levels. 
Although solid state array sensors are physically 
quantized in both space and intensity, the physical 
intensity (gray scale) quantization is not realized in 
the acquired digital image. The electron count in a 
sensor is converted to an amplified voltage signal by 
the camera circuitry. Besides the noise and distortion 
added to the intensity signal by the camera circuitry, 
the signal is usually further modified by filtering to a 
bandwidth of less than 5 MHz (about half the pixel 
rate). Under these conditions, a truely digital (piece- 
wise constant) image signal is never output. Note 
that this corruption of the raw array sensor data 
ocurrs in the camera and not in the imaging array 
itself. True "digital cameras" could be made which 
output a higher quality signal but commercially 
available cameras are designed according to image 
quality stardards set by the characteristics of human 
visual perception rather than the capabilities of 
digital image metrology. 
In principle, knowledge of the characteristics of the 
camera electronics would allow one to recover most 
of the raw image signal available at the chip but, 
unfortunately, the common use of the RS-170 video 
signal standard to transfer the image to a frame grab 
card (flash digitizer) ultimately eliminates such a 
possiblility. The RS-170 signal does not have 
provision for a synchronous pixel clock, thereby 
discarding most of the geometric integrity integral to 
the solid state imaging array. Without a synchronous 
clock, the digitizer must interpolate the position of 
each pixel between the start of successive line scans. 
Not only does the frame grab lack the necessary 
information to pin-point the timing of each pixel, it 
55