摘要 |
<P>PROBLEM TO BE SOLVED: To provide suitable and speedy image correction technology for outputting a suitable photographic print among photographic images taken under various photographic conditions. <P>SOLUTION: In the image correction technique, original images selected one after another out of many prepared sample images and sample proper images obtained from the original images through manual image corrections are compared and evaluated to generate a sample conversion table wherein correspondence relation is predetermined (#02), and a neural network is made to learn using a sample feature quantity set of a plurality of feature quantities, representing characteristics in the original images, as an input value to an input layer and a conversion table as an output value from an output layer, thereby correcting an input photographic image using the conversion table output from the feature quantity set computed from the input photographic image. <P>COPYRIGHT: (C)2008,JPO&INPIT |