Elsevier

Neural Networks

Volume 12, Issues 7–8, October–November 1999, Pages 1021-1036
Neural Networks

Computation of pattern invariance in brain-like structures

https://doi.org/10.1016/S0893-6080(99)00048-9Get rights and content

Abstract

A fundamental capacity of the perceptual systems and the brain in general is to deal with the novel and the unexpected. In vision, we can effortlessly recognize a familiar object under novel viewing conditions, or recognize a new object as a member of a familiar class, such as a house, a face, or a car. This ability to generalize and deal efficiently with novel stimuli has long been considered a challenging example of brain-like computation that proved extremely difficult to replicate in artificial systems. In this paper we present an approach to generalization and invariant recognition. We focus our discussion on the problem of invariance to position in the visual field, but also sketch how similar principles could apply to other domains.

The approach is based on the use of a large repertoire of partial generalizations that are built upon past experience. In the case of shift invariance, visual patterns are described as the conjunction of multiple overlapping image fragments. The invariance to the more primitive fragments is built into the system by past experience. Shift invariance of complex shapes is obtained from the invariance of their constituent fragments. We study by simulations aspects of this shift invariance method and then consider its extensions to invariant perception and classification by brain-like structures.

Section snippets

The problem of shift invariance

Our visual system can effortlessly recognize familiar objects despite large changes in their retinal images. The image of a given object changes due to variations in the viewing conditions, for example, changes in the viewing direction, illumination, position and distance. The visual system can somehow compensate for these changes and treat different images as representing an unchanging object. Many of the images we see are novel either because they depict objects not seen before, or because

Shift invariance by the conjunction of fragments

We have seen above the limitations of both the full replication and the single representation approaches to the problem of shift-invariant recognition. The full replication model is straightforward, and it uses the brain's inherent parallelism and the existence of multiple units responding selectively to a variety of different shapes. At the same time, the proposal to have a separate mechanism at each location tuned to each recognizable image is implausible because of its extreme redundancy and

Computation of pattern invariance in brain-like structures

In this section we first summarize the main properties of the approach to shift invariance and its implications, and then discuss the application of a similar approach to other aspects of invariant pattern perception.

Summary

Invariant perception is an achievement of biological visual systems that is difficult to replicate in artificial systems. We have outlined an approach to the computation of pattern invariance that appears more suitable for brain-like structures than alternative approaches. In this approach, invariance for complex patterns is based on a large number of stored relationships between more elementary image fragments. Invariant perception therefore depends on a continuous process of learning from

References (36)

  • I. Biederman

    Human image understanding: recent research and a theory

    Computer Vision, Graphics, and Image Processing

    (1985)
  • C.G. Gross et al.

    The neural basis of stimulus equivalence across retinal translation

  • A. Treisman et al.

    A feature integration theory of attention

    Cognitive Psychology

    (1980)
  • I. Biederman et al.

    Evidence for complete translational and reflectional invariance in visual object priming

    Perception

    (1991)
  • E. Bricolo et al.

    Translation-invariant features for recognition

    Perception

    (1992)
  • E. Bricolo et al.

    Further evidence for viewer-centered representation

    Perception

    (1993)
  • R. Desimone et al.

    Stimulus-selective properties of inferior temporal neurons in the macaque

    The Journal of Neuroscience

    (1984)
  • Dill, M., Edelman, S. (1997). Translation invariance in object recognition and its relation to other visual...
  • M. Dill et al.

    The role of visual field position in pattern-discrimination learning

    Proceedings of the Royal Society of London B

    (1997)
  • J. Fiser et al.

    Size invariance in visual object priming of gray-scale images

    Perception

    (1995)
  • P. Földiák

    Learning invariance from transformation sequences

    Neural Computation

    (1991)
  • K. Fukushima

    Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position

    Biological Cybernetics

    (1980)
  • J.L. Gallant et al.

    Selectivity for polar, hyperbolic, and cartesian gratings in macaque visual cortex

    Science

    (1993)
  • C.G. Gross

    Representation of visual stimuli in inferior temporal cortex

    Philosophical Transactions of the Royal Society of London B

    (1992)
  • D.H. Hubel et al.

    Receptive fields, binocular interaction and functional architecture in the cat's visual cortex

    Journal of Physiology

    (1962)
  • M. Ito et al.

    Size and position invariance of neuronal responses in monkey inferotemporal cortex

    Journal of Neurophysiology

    (1995)
  • P. Jolicoeur

    A size-congruency effect in memory for visual shape

    Memory and Cognition

    (1987)
  • Y. Le Cun et al.

    Back-propagation applied to handwritten zip code recognition

    Neural Computation

    (1989)
  • Cited by (0)

    View full text