| 摘要 |
<p>PURPOSE: To obtain a character recognition system with a low error ratio by quantizing the characteristics of plural fine line segments or microsegments constituting a mark by using a several dimensional cell-shaped array or many second-dimensional array-shaped arrays. CONSTITUTION: A data collecting device 10 supplies data describing a mark to a pre-processor 12 composed of an input buffer 20, output buffer 22, microprocessor 24, and memory 26. The pre-processor 12 maps the mark to be identified to the shape of an array, assigns a value to a cell constituting this array, and supplies this cell value to a neural network 14. The neural network 14 trained to identify or recognize the mark in a desired mark set identifies the target mark by using information supplied from the pre-processor 12. Also, the neural network 14 and the pre-processor 12 can be realized as a program in a computer, and the neural network 14 can be replaced with a vector quantizer.</p> |
