|
from problems of spurious peaks, as occuring with more conventional ridge detectors such as
Canny's. Owing to the hexagonal matrix, and the substantial blurring of the input image, upon which
the whole of our image processing is based, it has proved readily possible to provide a sequence of
logic steps which can result, automatically, in single, yet connected, profile strings. This obviates
much of the otherwise complex and inferential logic often necessary with peak detectors for profile
thinning and continuity.
A controlled test programme is described, which provided some measures of practical vernier
capabilities of our type of processing for both vernier position and vernier orientation, and for both
first and second difference approaches. Opportunity was taken to explore the capabilities both for
straight edges aligned at various orientations to the primary axes of the sampling matrix and for a
variety of disc stimuli. It is therefore believed that the total results demonstrate both the average
capabilities and the sensitivity of errors to interaction of geometrical profile location and sampling
matrix. In addition, since we use a hexagonal matrix processor, but are compelled frequently to use
images originally sampled on a square matrix as input material, we took the opportunity to carry out a
limited exploration of the errors and instabilities introduced by being forced to start with such square
matrix sampled images and then resample. A complete hardware system, of course, could, in
principle, carry out optical convolution and direct hexagonal matrix sampling.
|
