Having performed the image normalisation the 
segmentation of the image is possible using a single 
threshold value, as shown in Figure 4. 
0 8 . + ,. + 6 e 8 c» . 
e 9 8 9 9 e 9 9 e e 9 
° e e e e e 9 9 9 e 
e 
$ e 9» 9 a . 9 e 9 e e 
: 9 9 . 9 8 a 0 e e e 
e 
e 9 9 e 9 L © e 0 e e 
8 e e ê e e e e e e 8 
oe © ° ° a e © 9 —. e 9 
e 
e 9 $ 8 e 8 9 e e e 8 
e . + e 9 8 8 e © e 9 
Figure 4. Segmentation of original image. 
The improvements gained by using this method are only 
of benefit for initial target recognition as the possibility 
exists of a target appearing partially in more than one 
subimage and hence, being distorted. 
2.2 Detection and recognition of targets. 
The image normalisation discussed in the previous 
section compensates for the non-uniform background 
illumination and reflectivity. After image segmentation 
the process of searching for and recognising targets is 
performed. A single threshold value is chosen to create 
the binary image. In the case of image detection and 
recognition the prior knowledge of the size, shape and 
possible orientation of the targets is used. The process 
divides into two steps: (i) contour tracing of object, and 
(ii) extraction of a structure parameter, which includes 
area, perimeter and circle factor, to decide on the 
validity of targets. 
221 PA facite «T Targets It is necessary to 
trace the contour of all objects which appear in the 
binary image. There are many methods can be used for 
this purpose such as the chain code method (Pavlidis, 
1982). These are not discussed here as they are well 
understood and documented techniques. In the case of 
this study, the target appears as a black circular blob. 
The X,Y coordinates of the traced contour are 
extracted for use in analysis of the shape of the object. 
222 
iti When an image is segmented 
there may be many objects other than the legitimate 
targets, so that it is necessary to find a suitable method 
to distinguish between targets and non-targets. Typical 
features which can be used are (i) perimeter length, (ii) 
size, and (iii) shape. 
  
     
(i) Perimeter, The perimeter length of the subject can 
be calculated using the traced contour X,Y coordinates. 
(ii) Area, The area can be calculated by counting all of 
the pixels inside and on the perimeter of the subject. 
(iii) Shape. A shape factor is used to express the 
differences between circular subjects and non-circular. 
The definition of the shape factor is given in Equation 
2 
Q = A/[x (L/2)* ] (2) 
Where A is the area of the object, L is the longest 
distance across the object. The equation gives the ratio 
of the area of the subject to the area of a circle which 
circumscribes the subject. The nearer to a circle the 
object is, the closer to 1 the ratio is. The variation in the 
distance between pixels which are connected in the x,y 
and the pixels connected in diagonal directions has to 
be compensated for. Commercial cameras have 
differing scale factors in vertical and horizontal (a 
typical ratio for an industrial camera is 4:3 ). So there 
must also be an adjustment for this factor. 
These three factors, described by Equations 3,4, & 5, 
allow the building of a decision function which is able 
to establish the likelihood of a given subject being a 
target. All parameters used are expressed in relative 
values for convenience. 
Area factor = al. ABS((A-AA)/AA) (3) 
Perimeter factor = a2. ABS((P-PP)/PP) (4) 
Circle factor — a3.(1- Q) (5) 
Where: 
al,a2, and a3 are risk weight coefficients 
A is the estimated ideal target area 
P is the estimated ideal target perimeter 
Q is the circle factor 
AA is the actual area of the object 
PP is the perimeter of the object 
When all of these factors are within some 
predetermined bounds there is a high probability of this 
object being a target so the object coordinates are 
stored, otherwise the object is rejected. This process is 
repeated for all of the image until all of the objects are 
recognised as targets or rejected. Figure 5, shows the 
final results of the recognition process.