Digital Signal Processing Reference
In-Depth Information
Figure 19.5. Expanding circle
template applied to image of eye.
newly calculated origin and a reduced step size for the x and y
shifts.
We also see that the gradient is larger between the squares of
side length four and eight pixels than it is between the squares of
side length 8 and 12 pixels. This tells us which template squares
are more likely to border the target square. So, in each iteration of
the algorithm, as well as reducing the x and y step sizes, we would
also use the new best-candidate template-square size, and reduce
the step size of the expansion.
In the previous example we used a contour integral to deter-
mine the average gray level at each template position. This works
well if the region has uniform gray values but, if this is not the
case, it may be necessary to perform an area integral instead of
a contour integral. This has the effect of averaging the gray values
within the template area, and thus reducing the effect of image
details.
Once the target region has been located within the image, then
one of the focus algorithms previously described can be applied
to the region, and a focus score calculated.
Figure 19.5
shows the use of expanding circles to locate the
pupil within an image of the eye. The process is similar to the one
described for expanding and shifting square templates, but in this
case the template is a circle.
