422 
be 
  
423 
Key system elements, shown in Fig. 2, are: 
2,1. Image Capture: 
A 380 x 488 pixel CCD camera operating at a 7.16MH, clock rate is input to an 
8 bit 20MH, flash converter (Real, 1983). Standard 4MH, DRAM chips are used 
for storing the digitized image data at video rate. The output from this 
memory may be displayed directly upon a monitor after passing the data through 
a D/A converter. 
2.2. Fast Processor: 
This is an evolving sub system which presently consists of three identifiable 
parts; the spatial filter, dynamic filter coefficient delivery circuits and a 
correlator/convolver. 
(i) Spatial Operator: 
The quest for a two dimensional spatial operator which performs useful 
transformations rapidly with modest means has lead to processing by a nearest 
neighbour operator in the form of a convolution filter (Real, 1983). For 
economy, speed and sensitivity to high spatial frequencies, the 
implementation, Fig. 2, is in the form of a pair of 9 point filters in a 3 x 5 
configuration. They can be operated serially as independent filters or as a 
single pipelined unit at twice the throughput rate. If'a 3 x 3 block of 
pixels (with i, J as the central element) is selected for the convolution 
filter, then each emerging pixel is modified by its nearest neighbour as a 
finite convolution: 
1 1 
g(x,y) am 4 I A j l e f(x + i, y + 3 h(i, 13 (1) 
Excluding scaling and normalization, nine multiplications and eight addítions 
or subtractions are necessary for each pixel emerging from one 3x3 convolution 
filter. An image is therefore processed spatially and digitally by combining 
the digitized image data with a programmable m x n array point spread 
function. It is in general a non-linear operator the kernel of which has its 
elements individually set to any signed integer ín order to achieve a desired 
result. The filter stages are implemented with 100 ns. 8x8 
multiplier/accumulator arrays. A line memory and high speed data transfer 
register feeds digitized image data to one register of the filter input while 
the other receives the properly timed coefficient values.  Subsampling and 
time integration by frame feedback are options not indicated in Fig. 2. 
(ii) Coefficient Delivery: 
Rapid coefficient delivery is important in a dynamic or adaptive situation. 
For example, if the operator of a photogrammetric instrument is dissatisfied 
with the visibility of the images he sees, coefficient entry from keyboard as 
a large aerial photo is scanned would be most impractical. A coefficient 
scanner, Fig. 2, permits an image analyst to run through a líbrary of 
coefficient sets rapidly while viewing the results. A fast response is 
important because it is easier to optimize a process by subconscious human 
feedback when there is no interruption in the task flow. Keyboard entry is 
available if coefficient sets not prestored are required. 
Coefficients to be prestored are programmed into an E2PROM via keypad. Upon 
system power-up contents of the E2PROM are transferred to a high speed RAM for