By means of Gabor filter bank, establishing
partial image is computed. The algorithm for selection
of partial images having defective region is based on
different values achieved for variance. Naturally, the
image lying within the limit of our variance includes
defective region. In Fig. 1, right hand shows the
feature plans and left hand indicates related zoned
selected images. The resulted feature plan is
established by means of simple addition.
The average of results achieved from this
algorithm is shown in table 1. In this table, SNS is the
percent for correct detection of defective regions and
SPC is the same percent for non-defective regions.
The images on which experiments were conducted are
divided two groups. Group 1 includes images on
which the defects really exist. In group 2 there are
images on which we made the defects manually.
Results of both groups are offered. In showing the
results, partial images are considered non-normalized
and variances are normalized between 0and 1 by
dividing them in to their maximum value. The
threshold is considered 0.02 and the method used for
combination of partial images is simple addition