Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
GRAPHIC CONVERSION SYSTEM BY NEURAL NETWORK
Document Type and Number:
Japanese Patent JP3162154
Kind Code:
B2
Abstract:

PURPOSE: To easily generate a character font by causing a neural network to learn several graphics like characters having a certain title as samples.
CONSTITUTION: A neural network NN has a learning part A, and the weight of a weight update part 13 is used to evaluate unlearnt data after the end of learning in the learning part A. The learning part A is provided with a learning execution part 11 which inputs the input pattern of an input/output pattern holding part 10 for holding of input to the network NN and a teacher pattern as the output to the network NN to calculate the output value, a learning parameter holding part 12 where a prescribed learning rule and a leaning parameter are held, the weight update part 13 which compares the output of the network in the execution part 11 and the teacher pattern of the holding part 10 with each other based on the learning rule and the learning parameter to update the weight of the network, and a storage part 14 where the updated weight of the network is stored in an arbitrary learning stage.


More Like This:
Inventors:
Takehiko Tanaka
Masayuki Yokono
Application Number:
JP6378492A
Publication Date:
April 25, 2001
Filing Date:
March 19, 1992
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
富士通株式会社
International Classes:
G06G7/60; G06K9/66; G06K9/68; G06N3/00; G06N99/00; G06F15/18; G06T1/00; G09G5/24; (IPC1-7): G06G7/60; G06F15/18; G06K9/66; G06T1/00; G09G5/24
Domestic Patent References:
JP3189876A
JP3273761A
JP4109299A
Attorney, Agent or Firm:
Tsutomu Toyama (1 outside)