A neural model of contour integration in the primary visual cortex
Neural Computation
The time dimension for scene analysis
IEEE Transactions on Neural Networks
Unsupervised Segmentation With Dynamical Units
IEEE Transactions on Neural Networks
Hi-index | 0.00 |
We examine the dynamics of object recognition in a multi-layer network of oscillatory elements. The derivation of network dynamics is based on principles of sparse representation, and results in system behavior that achieves binding through phase synchronization. We examine the behavior of the network during recognition of objects with missing contours. We observe that certain network units respond to missing contours with reduced amplitude and temporal delay, similar to neuroscientific findings. Furthermore, these units maintain synchronization with a high-level object representation only in the presence of feedback.Our results suggest that the illusory contour phenomena are formal consequences of a system that dynamically solves the binding problem, and highlight the functional relevance of feedback connections.