Abstract
In the previous paper, the authors proposed a biological machine called Active Reticular Mechanism (abbreviated as ARM), and showed its excellent capabilities as a locomotive robot. The present paper deals with its another application as a pattern recognizer for curved surfaces, emphasizing the pattern recognition theory and the verification with experiments by using some pattern models.
Smooth and continuous surfaces of quadrics are considered as a set of categories to be recognized, so that spherical, ellipsoidal, hyperboloidal, cylindrical and planar surfaces are introduced. According to the property of a surface in the differential geometry, the chracteristics at a point on the surface are defind with a pair of principal curvatures, therefore the feature selections can be expressed by a pair of variables (ξ, η) derived from the curvatures.
The authors establish a specific theory of pattern recognition consisting of a set of linear inequalities laid on the ξ-η plane. In order to verify the clustering performance of the theory some experiments are carried out with the ARM model and the pattern models. As a result it is shown that the ARM can be utilized as a feature selecting observer by employing the authors' theory.
