unsharpness and use this as part of a knowledge base for interpretation.
On the other hand, it is inevitable that images recorded by imaging systems will be to some extent
unsharp, due to the spread function of the imaging optics. Such unsharpness will be at least that
defined by the Airy disc characteristics of imagery by a diffraction limited optical system, whilst it may
well be considerably greater than the Gaussian blur of S.D. = 1.3 sampling intervals recommended in
previous Chapters for near optimal information extraction by the sampling system. It is therefore
necessary to pose the question "Does it really matter how sharp edges are for analysis?". Indeed,
one should really enquire as to whether it matters if the input image is blurred or 'too sharp'.
Now, image sharpness can affect the analysis in two ways. Firstly, the sharpness may influence
the efficiency of edge sensing and extraction. Secondly, it may be desirable to be able to sense that
an image or individual edge is unsharp, and possibly to measure the extent of the unsharpness.
Such a capability could be particularly valuable, in analysis of otherwise pin sharp images, for the
detection and segmentation of solar shadow in outdoor scenes, (and also, of course, for other
shadow edges produced from extended light sources). It certainly is a well-established and readily
observable fact that human beings are usually instantly aware of which segmentable regions of
scenes are bounded entirely or partially by shadow edges. There is little doubt that one of the cues
permitting such recognition is the unsharpness of the edges. It is necessary, then, to question
whether we are in a position to measure unsharpness with our computer algorithms and whether, with
such knowledge, one needs to correct any of the computations described in earlier chapters.