Applications of Neural Networks in Micromechanics - Neural Networks and Micromechanics - page 193

Fig. 9.2 The position of the cutting tool relative to the workpiece

9.3 Permutation Coding Neural Classifier

A Permutative Coding Neural Classifier (PCNC) was developed for the recognition
of different types of images. It was proved on the MNIST database (handwritten
digit recognition) and the ORL database (face recognition) and showed good results
[ 11 , 12 ]. Here, we examine this classifier in micromechanical applications [ 13 ]. The
PCNC structure is presented in Fig. 9.3 . The image is input to the feature extractor.
The extracted features are applied to the encoder input. The encoder produces
an output binary vector of large dimension, which is presented to the input of the
one-layer neural classifier. The classifier output gives the recognized class.

9.3.1 Feature Extractor

An initial image (Fig. 9.4 ) is to be input to the feature extractor. The feature
extractor begins with the selection of specific points in the image. In principle,
various methods of selecting the specific points could be proposed. For example,
the contour points could be selected as the specific points. The rectangle of h l w size
is formed around each specific point (Fig. 9.5 ).
The specific points were selected using the following procedure. For each set of
four neighboring pixels, we calculate the following expressions:

;

d 1 ¼ br ij br iþ 1 jþ 1

;

d 2 ¼ br ijþ 1 br iþ 1 j (9.1)

D ¼ max ð d 1 ; d 2 Þ;