Robotics Reference
In-Depth Information
Figure 28.
A fish, showing the image of its eye and the numerical representation of that
image (from the web site “Computer Vision” by James West)
hand corner of the image there is an even lighter shade, corresponding
to the number 17. Note also that some pixels have higher numbers than
one might expect from this simplistic example—the reason being that
each number here is the
code
for a different shade, and it is not always
the case that a higher number means a darker shade.
In a black-and-white image the lowest possible number (usually 0)
might correspond to white, with the highest number meaning black
and the in-between numbers corresponding to different shades of grey.
In a colour image the numbering system is more complex because the
numbers not only represent how light or dark a pixel is, they also indi-
cate the colour of the pixel (and the number of possible colours can be
enormous).
Printed Character Recognition
One of the earliest vision problems to be subjected to machine recogni-
tion technology was the recognition of hand-written characters. In fact
this is a more difficult task than dealing with printed characters so we
shall discuss printed character recognition first.
Most early character recognition systems consisted of two main proc-
essing units—a character separator to identify each unit that needed to be
recognized, and the isolated character recogniser itself. Character sepa-
ration (also called segmentation) is important because printed characters
can be of different sizes and can be separated from their neighbour char-
acters by different distances, so care is often needed to avoid confusing
part of one character with an adjacent character. Most printed character
recognition systems operate in either a fixed spacing mode, where the
sizes of the characters are known in advance and therefore the segmenta-
tion process can be very reliable, or in a variable spacing mode, where no
advance information can be assumed about the sizes of the characters.
