Perception Training
The perception training is the problem of finding if is discriminate function find by the perception training shown in Fig.7.
Fig.7 The perception training
A vector W that correctly separates all the training patterns. i.e.
is the discriminate function where are augmented training patterns, that is , if is a training pattern, from class . where Xj is a training pattern from is a (d+1) vector.
For example, if there are five 1-d patterns distributed on the feature space (Fig.8):
{2, 3, 4} : three training patterns from
{
-1,-0.5}: two training patterns from
Fig.8 the training patterns
Then we can obtain the augment training pattern:
as show in Fig.9.
Fig.9 the augmented pattern space
we can see that:
Fig.10 The weighting space
In Fig.10 we can obtain the allowable area of W. We can choose a suitable W to discriminate all patterns into groups of object and background[4]. Here we introduce a 7_7 discriminate mask operation as follows (Fig.11):
Fig.11 a 7 X 7 discriminate mask operation
1.notation
ROI: region of interest in LCD plate.
T1:the test region of a discriminate function.
g(.): to detect are there any spacers in T1
mark(.): to label the pixel
2.initialization
max=0 ,min= 255
3.recursion
repeat
for all
find out the maximum and minimum value of T1.
if g (max-min)==true
mark();
else
let T1 to next position
until T1 out of ROI