This paper deals with a newly developed methodology and a related system for the automatic pattern recognition of machine parts, etc., on the basis of the degree of similarity. In the methodology, two feature parameters are extracted. These two parameters are the pattern-classification value
S and the group-classification value
N. The pattern-classification value is obtained from the ratio of the area to the product of the largest width multiplied by the largest height of the projection of the pattern on the reference line. The group-classification value is obtained by the second differentiation of the projection. Since the patternclassification value generally varies with the postural angle of the shape, the average value
S and the range of variance Δ
S of the pattern-classification value are defined.
A pattern is mapped firstly on the feature extraction space (named
E-space) decided by using three parameters
S, Δ
S and
N. Secondly, the
E-space is mapped on the sensing filter space (
F-space). The similarity is calculated by the distance between the two patterns mapped on the
F-space. In order to match the calculated similarity with the similarity recognized by human, the characteristics shown in the similarity recognition of various kinds of shapes have been examined by a few adults and the experimental results are taken into account in developing the system. By the developed pattern recognition system, one pattern can be recognized in less than 4 seconds to be classified according to the 22 kinds of matching patterns memorized in the minicomputer.
抄録全体を表示