| 摘要 |
PURPOSE: To eliminate the difficulty in preparing a threshold value to prevent a mark from being misjudged as a similar mark in a mark recognizing device. CONSTITUTION: Image data obtained by picking up the image of the mark of a learing object is made to degenerate to a multidimensional input signal vector, and to which it is closest among prestored plural reference vectors is detected by a vector quantization technology. The closest reference vector is renewed so as to be further approached to the input signal vector, and learning is performed. The quality of the mark is judged by using a reference vector after learning. The quality is judged by comparing the distance between the input signal vector obtained from the mark and the reference vector after learning with a threshold value. Since the quality of the mark is judged after learning is performed on the pattern of a representative mark, threshold value adjusting work can be simplified. |
