Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

A hybrid learning network for shift, orientation, and scaling invariant pattern recognition.

R Wang1

  • 1Engineering Department, Harvey Mudd College, Claremont, CA 91711, USA. ruye_wang@hmc.edu

Network (Bristol, England)
|January 5, 2002
PubMed
Summary

This study introduces a novel three-layer neural network for invariant visual pattern recognition. The network effectively identifies patterns regardless of their geometric variations like size and orientation.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

[Osteoarthropathy in pregnancy].

Zhonghua fu chan ke za zhi·2001
Same author

Glomerular laminin isoform transitions: errors in metanephric culture are corrected by grafting.

American journal of physiology. Renal physiology·2001
Same author

[The relationship of vascular endothelial growth factor and angiogenesis to the progression of gastric carcinoma].

Zhonghua bing li xue za zhi = Chinese journal of pathology·2001
Same author

Monte Carlo dose calculations of beta-emitting sources for intravascular brachytherapy: a comparison between EGS4, EGSnrc, and MCNP.

Medical physics·2001
Same author

Tmtacn, tacn, and triammine complexes of (eta 6-arene)OsII: syntheses, characterizations, and photosubstitution reactions (tmtacn = 1,4,7-trimethyl-1,4,7-triazacyclononane; tacn = 1,4,7-triazacyclononane).

Inorganic chemistry·2001
Same author

A panel immunoblot using co-incubated monoclonal antibodies for identification of melanoma cells.

Journal of immunological methods·2001

Area of Science:

  • Computational Neuroscience
  • Artificial Intelligence
  • Machine Learning

Background:

  • Visual pattern recognition is a fundamental cognitive function.
  • Achieving invariance to geometric transformations (translation, rotation, scale) remains a challenge in artificial systems.
  • Biological visual systems exhibit remarkable robustness to such variations.

Purpose of the Study:

  • To propose a generic and biologically plausible neural network model for invariant visual pattern recognition.
  • To develop a hybrid learning algorithm that enables geometric invariance.
  • To demonstrate the network's potential as a computational model for biological object recognition.

Main Methods:

  • A three-layer neural network architecture was designed.

Related Experiment Videos

  • A hybrid learning algorithm combining competitive and Hebbian learning was employed.
  • Internal representation of geometric variations was achieved through lateral connections between input and middle layers.

    • Main Results:

    • The network successfully achieved pattern recognition invariant to translation, orientation, and scale.
    • Each pattern was represented by a specialized subset of middle-layer nodes.
    • Output layer nodes were trained to recognize patterns independent of their geometric appearance.

    Conclusions:

    • The proposed neural network offers a robust and generic approach to invariant visual pattern recognition.
    • The hybrid learning mechanism effectively handles geometric variations.
    • The model provides insights into the mechanisms of invariant object recognition in biological visual systems.