| EPSRC Reference: |
GR/S81339/01 |
| Title: |
AIBACS: Computational Modelling of Top-down Effects in Visual Information Processing |
| Principal Investigator: |
Professor MH Johnson |
| Other Investigators: |
|
| Researcher Co-investigators: |
|
| Project Partners: |
|
| Department: |
Psychological Sciences |
| Organisation: |
Birkbeck College |
| Scheme: |
Standard Research |
| Starts: |
01 October 2004 |
Ends: |
30 September 2007 |
Value (£): |
206,373
|
| EPSRC Research Topic Classifications: |
| Information and communication technologies: Vision, Hearing and Other Senses - Applications in ICT |
Medical and health interface: Biomedical neuroscience |
|
| EPSRC Industrial Sector Classifications: |
|
| Related Grants: |
|
| Panel History: |
|
|
Summary |
|
The cerebral cortex is the brain structure responsible for all high-level mental capacities such as reasoning, memory, language, and perception. Understanding how the cortex functions is thus crucial to understanding human and animal intelligence and for developing biologically inspired machines that can perform intelligent tasks. Visual perception is accomplished by a number of distinct areas of the cortex which specialise in processing different aspects of the visual stimulus. The specific functions performed by these cortical areas have been extensively studied. However, cortical areas do not operate in isolation. Each area is influenced by the activity of other areas, and the overall task of seeing is achieved through specialised cortical areas cooperating with, and constraining, each other in order to produce a consistent interpretation of the world. While such coordinated activity is fundamental to many of our perceptual abilities, the mechanisms by which it is achieved are poorly understood. This project aims to investigate the mechanisms by which cortical regions interact by developing a computational simulation that can be used to explore this issue. It is hoped that this research will significantly advance our understanding of many perceptual processes, and in so doing, make it possible to build improved artificial vision systems. Hence, this project aims to combine the study of computational and biological systems to the mutual benefit of both disciplines: by employing computational techniques to aid our understanding of human visual perception and by extracting computational principles from the biological system that will be useful for developing artificial vision systems.
|
| Final Report Summary |
This project has explored the brain mechanisms underlying visual perception. A computational model has been developed which is consistent with the anatomy and physiology of the cortex and which can explain a number of perceptual phenomena. Additional behavioural experiments were also performed to help constrain the model. The principal outcomes of this research were as follows:
1. A more complete model of the image segmentation process has been developed. In contrast to previous work, this model can explain the influence of various factors on the outcome of figure-ground segmentation including both exogenous factors (image properties such as Gestalt cues) and endogenous factors (top-down influences such as prior knowledge and attention).
2.It has been demonstrated, behaviourally, for the first time that recently learnt object knowledge can influence subsequent figure-ground segmentation.
3. It has been demonstrated, for the first time, that attention can influence Gestalt cues to image segmentation.
4. A novel biophysical mechanism has been proposed via which cortical lateral connections and cortical feedback connections interact. It has been shown that this mechanism can explain a series of psychophysical phenomena concerned with collinear facilitation.
5. A novel mechanism has been proposed for learning object representations that are invariant to viewpoint,. In contrast to previous work in this area, the proposed mechanism can work in more realistic situations where multiple objects are present in the scene and hence is more appropriate for real-world computer vision applications.
6. It has been shown that two highly influential theories of visual perception (predictive coding and biased competition) can be reconciled. This insight has significant implications for theoretical neuroscience by providing a unified theory, and provides a solid theoretical basis for the proposed model.
|
| Further Information: |
|
| Organisation Website: |
http://www.bbk.ac.uk/ |