| 摘要 |
An image recognizing method is provided, which includes the capability of efficiently recognizing characters such as letters and numerical characters included in an original image with accuracy when at least one of the characters is composed of plural elements. In this method, the elements in the original image are extracted to obtain a second image, in which each of the elements is enclosed by a rectangular frame. Then, a composite image is prepared with respect to a combination of the rectangular frames in the second image. After calculating a characteristic amount of the elements included in the composite image, the characteristic amount is input in a back-propagation network, in which learning about a reference character(s) to be included in the image has already finished, so that a degree of agreement between the characteristic amount of the composite image and the reference character is obtained. The composite image having a highest degree of agreement therebetween is determined from results provided by the back-propagation network with respect to different combinations of the rectangular frames in the second image, to output it as a recognition data.
|
