Title:
【発明の名称】階層型ニューラルネットワークを用いた物体認識方式
Document Type and Number:
Japanese Patent JP3110167
Kind Code:
B2
Abstract:
PURPOSE:To resolve the optimizing problem for searching an object model and conversion maximizing the degree of any measurement between the image of an object and the model by using a hierarchical neural network concerning the object recognizing system for specifying the shape or position of the object or any category. CONSTITUTION:This system is composed of a learning process 11 for applying data as the reocgnized result to an input layer unit and applying sensor information concerning the object to an output layer unit (A), and recognizing process 12 for searching the resolution, which reduces conflict between the output of the output layer unit and the sensor information and satisfies limit conditions concerning the external field, by changing the input to the input layer unit and minimizing an energy function corresponding to the conflict and the limit conditions (B).
Inventors:
Daiki Masumoto
Takashi Kimoto
Shigemi Nagata
Takashi Kimoto
Shigemi Nagata
Application Number:
JP24624392A
Publication Date:
November 20, 2000
Filing Date:
September 16, 1992
Export Citation:
Assignee:
富士通株式会社
International Classes:
G06F15/18; G06G7/60; G06N3/00; G06N3/04; G06N99/00; G06T7/00; (IPC1-7): G06T7/00; G06G7/60; G06N3/00; G06N3/04
Attorney, Agent or Firm:
Yoshiyuki Osuge (1 outside)