PURPOSE: To improve the efficiency and the precision of character recognition by using a hierarchized network which has several layers of restrictive feature detection each of which includes plural restrictive feature maps and plural feature reduction maps corresponding to these maps.
CONSTITUTION: Respective arrays (boxes) presented in levels 20 to 60 include plural calculation elements per array unit. The first feature detection layer includes plural restrictive feature maps 20 and plural feature reduction maps 30 corresponding to maps 20, and the second feature detection layer includes plural restrictive feature maps 40 and plural feature reduction maps 50 corresponding to maps 40, and the lasy layer of the network includes a character classification layer 60 completely connected to all feature reduction maps of the second feature detection layer. The network recognizes characters from a supplied image through a large quantity of parallel calculation. Thus, characters are automatically efficiently recognized with a high reliability.
RICHIYAADO II HAWAADO
ROORENSU DEI RATSUKERU
YAN REKAN
