| 摘要 |
<p>The present invention provides a method useful in image recognition by employing the restoration of degraded images to produce an approximation of the original ideal artwork for the image. That is, pursuant to the invention, a set of images is read from an image source and an attempt is made to recover, or closely approximate, the ideal artwork (e.g., original symbol) from which the images were derived. The approximation of the original symbol then serves to train an imageclassifier thereby increasing the effectiveness of the classifier. For example, the classifier can use the original symbol approximations as input to a degradation model to produce a set of degraded images useful in document classification. Thus, the effectiveness of the classifier is enhanced through the reduction in the classification error rate.</p> |
