This paper describes initial work on an image analysis
method designed for use in
automatic detection of early symptoms of
diabetic retinopathy (DB). DB is an eye disease
which appears as a bieffect of diabetes.
The earliest symptom is the
appearance of small dark spots, termed microaneurysms (MA),
on the retina. Both
their number and placement has diagnostic importance.
It is therefore of great interest if an automatic algorithm for detecting
MA from retinal images can be found.
However, there are many difficulties in doing this,
such as a nonuniform image background, various kinds of noise,
and the presence of blood vessels --- some of which, due to local intensity
variations, will appear to have nonconnected endings.
In our work we have combined median filtering and morphological operators
in a method that takes at least some of these detractions into account.
The problem so far is that,
although the algorithm detects most of the MA, it also detects quite
a lot of false spots stemming from imperfections in the imaging process.
However, this may not be such a problem if higher quality image material
were to be used.