Introduction - Feature Extraction and Image Processing - page 8

sensors are combined prior to transmission through the optic nerve. This is an experimental
model, since there are many ways possible to combine the different signals together. For
further information on retinal neural networks, see Ratliff (1965); an alternative study of
neural processing can be found in Overington (1992).

1.3.3 Processing

The neural signals are then transmitted to two areas of the brain for further processing.
These areas are the associative cortex , where links between objects are made, and the
occipital cortex , where patterns are processed. It is naturally difficult to determine precisely
what happens in this region of the brain. To date, there have been no volunteers for detailed
study of their brain's function (though progress with new imaging modalities such as
Positive Emission Tomography or Electrical Impedance Tomography will doubtless help).
For this reason, there are only psychological models to suggest how this region of the brain
operates.
It is well known that one function of the eye is to use edges, or boundaries , of objects.
We can easily read the word in Figure 1.6 (a), this is achieved by filling in the missing
boundaries in the knowledge that the pattern most likely represents a printed word. But we
can infer more about this image; there is a suggestion of illumination, causing shadows to
appear in unlit areas. If the light source is bright, then the image will be washed out,
causing the disappearance of the boundaries which are interpolated by our eyes. So there
is more than just physical response, there is also knowledge, including prior knowledge of
solid geometry. This situation is illustrated in Figure 1.6 (b) that could represent three
'Pacmen' about to collide, or a white triangle placed on top of three black circles. Either
situation is possible.

(a) Word? (b) Pacmen?

Figure 1.6 How human vision uses edges

It is also possible to deceive the eye, primarily by imposing a scene that it has not been
trained to handle. In the famous Zollner illusion , Figure 1.7 (a), the bars appear to be
slanted , whereas in reality they are vertical (check this by placing a pen between the lines):
the small crossbars mislead your eye into perceiving the vertical bars as slanting. In the
Ebbinghaus illusion , Figure 1.7 (b), the inner circle appears to be larger when surrounded
by small circles, than it appears when surrounded by larger circles.