Related Experiment Videos
A multi-layer sparse coding network learns contour coding from natural images.
Patrik O Hoyer1, Aapo Hyvärinen
1Neural Networks Research Centre, Helsinki University of Technology, P.O. Box 9800, FIN-02015 HUT, Finland. patrik.hoyer@hut.fi
Vision Research
|June 21, 2002
Summary
This study proposes a model where complex cell responses in the visual system are sparsely represented, leading to contour coding and end-stopped receptive fields for efficient image processing.
Area of Science:
- Visual neuroscience
- Computational neuroscience
- Image processing
Background:
- Early visual system function is linked to natural input statistics.
- Primary visual cortex properties like receptive fields and topography reflect efficient coding of natural images.
Purpose of the Study:
- Extend efficient coding framework to complex cell responses.
- Investigate sparse representation in higher-order neural layers.
- Model contour coding and end-stopped receptive fields.
Main Methods:
- Computational modeling of neural responses.
- Analysis of efficient coding principles.
- Framework extension to higher-order visual processing.
Main Results:
- Complex cell responses can be sparsely represented by a higher-order layer.
- This sparse representation explains contour coding and end-stopped receptive fields.
- Contour integration is modeled as top-down inference.
Conclusions:
- Sparse coding provides a unified framework for early visual processing.
- The model offers insights into contour perception and neural computation.
- Efficient coding principles extend to complex cell functions and visual perception.