Email updates

Keep up to date with the latest news and content from BMC Neuroscience and BioMed Central.

This article is part of the supplement: Eighteenth Annual Computational Neuroscience Meeting: CNS*2009

Open Access Poster presentation

Feedback in a hierarchical model of object recognition in cortex

Salvador Dura*, Thomas Wennekers and Sue Denham

Author Affiliations

Centre for Theoretical and Computational Neuroscience, University of Plymouth, Plymouth, PL4 8AA, UK

For all author emails, please log on.

BMC Neuroscience 2009, 10(Suppl 1):P355  doi:10.1186/1471-2202-10-S1-P355

The electronic version of this article is the complete one and can be found online at: http://www.biomedcentral.com/1471-2202/10/S1/P355


Published:13 July 2009

© 2009 Dura et al; licensee BioMed Central Ltd.

Poster presentation

Substantial evidence exists showing V1/V2 cortical activity in response to illusory contours (e.g. Kanizsa's triangle)[1]. V1 represents information that correlates with visual perception but does not arise from the ordinary feedforward pathways (retina and LGN), suggesting this response is driven by feedback connections from higher-level areas. At the same time, several authors have emphasized the important role played by feedback connections in visual perception and have suggested hierarchical Bayesian inference as a plausible underlying interpretation [2-5]. The proposed model aims at providing a functional interpretation of the illusory contour phenomenon and, more generally, of the role of feedback connections in visual perception under the perspective of a Bayesian inference framework.

The model takes as a starting point the HMAX model, a well-known cortex-based object recognition model [6,7], and shows how this feedforward architecture can also be interpreted in probabilistic terms and extended to include feedback using the Bayesian Belief Propagation algorithm. Feedforward responses are understood as probability distributions of the set of features at the different positions, which build up hierarchically obtaining at the top level the most probable object represented by the input image. Activity relating to the stored invariant prototype of this object is then fed back down the network and combined with the feedforward response at each level to obtain the resulting belief. However, high-level invariant object prototypes generate ambiguous and diffuse feedback that needs refinement. The disambiguation process, based on cues from existing feedforward responses, uses a local extrapolation algorithm that exploits co-linearity, co-orientation and good continuation to guide and adapt feedback.

Simulation results show successful feedforward object recognition, including cases of occluded and illusory images. Recognition is both position and size invariant. The model also provides a functional interpretation of feedback connectivity which accounts for several observed phenomena. Model responses qualitatively match representations in early visual cortex of occluded and illusory contours [1]; and fMRI data showing that high-level object recognition reduces activity in lower areas [8]. Additionally a dynamic mechanism for illusory contour formation is proposed which can adapt a single high-level object prototype to Kanizsa's illusory figures of different sizes, shapes and positions. The mechanism is supported by experimental studies that suggest the interaction between contextual feedback signals and local evidence in V1 is responsible for contour completion [1,9]. By imposing top-level priors the model can also simulate the effects of spatial attention, priming and visual search.

References

  1. T Nguyen M: Dynamics of subjective contour formation in the early visual cortex.

    Proc Natl Acad Sci USA 2001, 98:1907-1911. PubMed Abstract | Publisher Full Text | PubMed Central Full Text OpenURL

  2. Friston K: Learning and inference in the brain.

    Neural Netw 2003, 16:1325-1352. PubMed Abstract | Publisher Full Text OpenURL

  3. Lee TS, Mumford D: Hierarchical Bayesian inference in the visual cortex.

    J Opt Soc Am A Opt Image Sci Vis 2003, 20:1434-1448. PubMed Abstract | Publisher Full Text OpenURL

  4. Rao RPN, Ballard DH: Dynamic model of visual recognition predicts neural response properties in the visual cortex.

    Neural Computat 1997, 9:721-763. Publisher Full Text OpenURL

  5. Sillito AM, Cudeiro J, Jones HE: Always returning: feedback and sensory processing in visual cortex and thalamus.

    Trends Neurosci 2006, 29:307-316. PubMed Abstract | Publisher Full Text OpenURL

  6. Cadieu C, Kouh M, Pasupathy A, Connor CE, Riesenhuber M, Poggio T: A model of V4 shape selectivity and invariance.

    J Neurophysiol 2007, 98:1733-1750. PubMed Abstract | Publisher Full Text OpenURL

  7. Riesenhuber M, Poggio T: Hierarchical models of object recognition in cortex.

    Nat Neurosci 1999, 2:1019-1025. PubMed Abstract | Publisher Full Text OpenURL

  8. Murray SO, Schrater P, Kersten D: Perceptual grouping and the interactions between visual cortical areas.

    Neural Netw 2004, 17:695-705. PubMed Abstract | Publisher Full Text OpenURL

  9. Stanley DA, Rubin N: fMRI Activation in Response to Illusory Contours and Salient Regions in the Human Lateral Occipital Complex.

    Neuron 2003, 37:323-331. PubMed Abstract | Publisher Full Text OpenURL