Abstract
In this paper, an approach to recognize a natural shape or a planar picture written with one stroke is treated.
It is proposed to use the shape vector to describe natural shapes excluding symbols such as alphabeticals or numericals. The shape vector is defined as a vector composed with the series of external angles of a shape considered as an equilateral polygon. The searching shape set is introduced as a category set to be recognized. And the shape distance is defined as a measure to judge the similarity of a pair of shapes. It corresponds to the minimum Euclidean distance between the points coordinate to the pair of shapes in an n-dimensional space.
In order to verify that the shape distance is equivalent to a human impression with respect to the similarity among the various shapes, an experiment on a digital computer has been carried out. In the experiment, shape vectors are prepared from seventy-nine arbitrarily hand-written pictures drawn with one stroke. Fifteen searching shapes are selected from the pictures, and a great deal of computation of shape distances is performed for every searching shape to the remaining reference pictures. Additional computation with another measurement, a kind of correlation coefficient, proposed by Dr. T. Iijima is also performed to compare it with the authors' result.
The experiment shows good agreement with the human impression.
