The gray level of the image’s background of a LCD plate will vary as we inspect different LCD plates. Thus we can’t use a fix threshold value to separate the spacers and the background into two groups[2]. We therefore design a discriminate function to separate the training patterns from two pattern classes when nothing is known about the pattern class distributions.

Let state space Ω be defined as

The action space A can be define as

Here we have two truth classes and two man made classes .Suppose for pattern .Then we can find a discriminate function as follows:

g ( . ) for which

g ( xi ) > 0 if θi = 1 .

g ( xi ) < 0 if θi = 2 .

In Fig.4 we create two classes which distribute on the diagram in fig.4 with 6 patterns for each class to draw the curve . There is a critical area between the curve of and . We can get the discriminate function by training data in this range[3].

Discriminate function in general from:

Where Parameters W

is from the training patterns.

The flow chart (Fig.5) shows the scheme method of separating prearranged data into two classes.

index