algorithm have been developed to overcome this problem. 
Widrow discussed the idea of using "rubber mask" templates, 
in which the Shape and size of the template could vary /6/. 
The basic template would be matched over and over, in 
different configurations, until the best match was achieved. 
Ackermann used a transformation approach in which a 16 x 16 
Or 32 x 32 window was employed to correlate imagery between 
two members of a Stereopair I The windows underwent a 
least squares affine transformation to correct for the 
effects of perspective distortion and relief displacement. 
The correlation Of features on sequential x-ray images 
presents several additional problems: As features in the 
cranial region develop, the X-ray images change in two ways. 
Individual features change in Shape, size, and orientation. 
However, because the X-ray image consists of a projection of 
all radiopaque surfaces in the head onto a single plane, the 
background "noise" caused by the shift in underlying and 
overlying features also changes. 
Therefore, a different matching technique was developed for 
anatomic feature tracking over multiple images. On the 
baseline image the feature, as identified by the operator, 
Was stored as a digital sub-image. The sub-image was 
enhanced by convolving it with a strongly intensifying 
Laplacian filter: 
=3 -2 -2 
-2 16 -2 
-2 -2 -2 
A sample feature output is shown in Figure 2. The filter, 
in combination with the characteristics of the x-ray image, 
Served to reduce the features in the image to outlines. 
The boundary of the outline was then automatically extracted 
as row and column Pixel numbers. an algorithm was developed 
to "walk" around the feature boundary, deciding on the most 
likely outline coordinates. Decisions in the:algorithm were 
based on the direction and intensity of the density gradient 
at each point. It also was programmed to avoid sharp 
changes in direction, thus Smoothing the outline somewhat. 
If the boundary ended at any point, a search was performed 
in the immediate neighbourhood in order to determine if the 
-. 166 -