Kunihiko Fukushima

ORCID: 0000-0003-3025-7627
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Neural Networks and Applications
  • Neural dynamics and brain function
  • Visual perception and processing mechanisms
  • Advanced Vision and Imaging
  • Advanced Memory and Neural Computing
  • Image Processing and 3D Reconstruction
  • Image Retrieval and Classification Techniques
  • CCD and CMOS Imaging Sensors
  • Handwritten Text Recognition Techniques
  • Image Processing Techniques and Applications
  • Color Science and Applications
  • Nuclear Physics and Applications
  • Neural Networks and Reservoir Computing
  • Retinal Development and Disorders
  • Infrared Target Detection Methodologies
  • Blind Source Separation Techniques
  • Industrial Vision Systems and Defect Detection
  • Hand Gesture Recognition Systems
  • Face and Expression Recognition
  • Advanced Image and Video Retrieval Techniques
  • Image and Signal Denoising Methods
  • Atomic and Subatomic Physics Research
  • Spatial Cognition and Navigation
  • Underwater Acoustics Research
  • Robotics and Automated Systems

Fuzzy Systems Institute
2012-2024

Osaka University
1994-2023

Mitsubishi Electric (Japan)
2016

Kansai University
2006-2012

Tokyo University of Technology
2001-2006

University of Electro-Communications
1999-2003

Kyoto University
1988-1995

Family Inada (Japan)
1994

Kyoto Bunkyo University
1990

NHK Spring (Japan)
1988

A recognition with a large-scale network is simulated on PDP-11/34 minicomputer and shown to have great capability for visual pattern recognition. The model consists of nine layers cells. authors demonstrate that the can be trained recognize handwritten Arabic numerals even considerable deformations in shape. learning-with-a-teacher process used reinforcement modifiable synapses new model, instead learning-without-a-teacher applied previous model. focus mechanism rather than self-organization.

10.1109/tsmc.1983.6313076 article EN IEEE Transactions on Systems Man and Cybernetics 1983-09-01

A model of a neural network is presented that offers insight into the brain's complex mechanisms as well design principles for information processors. The has properties and abilities most modern computers pattern recognizers do not possess; recognition, selective attention, segmentation, associative recall. When composite stimulus consisting two or more patterns presented, pays attention to each one after other, segments from rest, recognizes it separately in contrast earlier models. This...

10.1109/2.32 article EN Computer 1988-03-01

A pattern recognition system which works with the mechanism of neocognitron, a neural network model for deformation-invariant visual recognition, is discussed. The neocognition was developed by Fukushima (1980). has been trained to recognize 35 handwritten alphanumeric characters. ability deformed characters correctly depends strongly on choice training set. Some techniques selecting patterns useful large number are suggested.

10.1109/72.97912 article EN IEEE Transactions on Neural Networks 1991-05-01

A new type of visual feature extracting network has been synthesized, and the response simulated on a digital computer. This research done as first step towards realization recognizer handwritten characters. The design was suggested by biological systems, especially, systems cat monkey. is composed analog threshold elements connected in layers. Each element receives inputs from large number neighbouring layers performs its own special functions. It takes care one restricted part...

10.1109/tssc.1969.300225 article EN IEEE Transactions on Systems Science and Cybernetics 1969-01-01

A neural network model of selective attention is discussed. When two patterns or more are presented simultaneously, the successively pays to each one, segmenting it from rest and recognizing separately. In presence noise defects, can recall complete pattern in which has been eliminated defects corrected. These operations be successfully performed regardless deformation input patterns. This an improved version earlier proposed by author: ability segmentation lateral inhibition.

10.1364/ao.26.004985 article EN Applied Optics 1987-12-01

10.1016/s0925-2312(02)00614-8 article EN Neurocomputing 2003-03-15

10.1007/bf00272461 article EN Biological Cybernetics 1973-02-01

10.1016/j.neunet.2013.01.001 article EN Neural Networks 2013-01-14

An electronic model of the retina has been developed with a new technique in making interconnections between cells. The consists about 700 photoreceptors and same number output cells have concentric on-center type receptive fields show Mach-band Broca-Sulzer effects.

10.1109/proc.1970.8066 article EN Proceedings of the IEEE 1970-01-01

10.4249/scholarpedia.1717 article cc-by-nc-sa Scholarpedia 2007-01-01

Deep convolutional neural networks (deep CNNs) show a large power for robust recognition of visual patterns. The neocognitron, which was first proposed by Fukushima (1979), is recognized as the origin deep CNNs. Its architecture suggested neurophysiological findings on systems mammals. It acquires ability to recognize patterns robustly through learning. Although neocognitron has long history, improvements network are still continuing. This article discusses recent focusing differences from...

10.1109/tsmc.2020.3042785 article EN IEEE Transactions on Systems Man and Cybernetics Systems 2021-01-01
Coming Soon ...