Page 26 - Classification Parameter Estimation & State Estimation An Engg Approach Using MATLAB

DETECTION AND CLASSIFICATION 15

Example 2.1 Classification of small mechanical parts
Many workrooms have a spare part box where small, obsolete
mechanical parts such as bolts, rings, nuts and screws are kept. Often,
it is difficult to find a particular part. We would like to have the parts
sorted out. For automated sorting we have to classify the objects by
measuring some properties of each individual object. Then, based on
the measurements we decide to what class that object belongs.
As an example, Figure 2.2(a) shows an image with rings, nuts, bolts
and remaining parts, called scrap. These four types of objects will be
classified by means of two types of shape measurements. The first
type expresses to what extent the object is six-fold rotational sym-
metric. The second type of measurement is the eccentricity of the
object. The image-processing technique that is needed to obtain these
measurements is a topic that is outside the scope of this book.
The 2D measurement vector of an object can be depicted as a point
in the 2D measurement space. Figure 2.2(b) shows the graph of the
points of all objects. Since the objects in Figure 2.2(a) are already
sorted manually, it is easy here to mark each point with a symbol that
indicates the true class of the corresponding object. Such a graph is
called a scatter diagram of the data set.
The measure for six-fold rotational symmetry is suitable to discrim-
inate between rings and nuts since rings and nuts have a similar shape
except for the six-fold rotational symmetry of a nut. The measure for
eccentricity is suitable to discriminate bolts from the nuts and the rings.

(a) (b) 1
bolts
nuts
0.8
rings
measure of eccentricity 0.6
scrap

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
measure of six-fold rotational symmetry

Figure 2.2 Classification of mechanical parts. (a) Image of various objects,
(b) Scatter diagram