How the visual system can detect feature homogeneity from spike latencies

We propose that homogeneous image regions may be used to quickly segment a visual scene, when an object hypothesis is not yet available. In a cooperative model of the konio-cellular and the parvo- or magno-cellular visual pathway of primates, we demonstrate how homogeneity-selective responses can be gained from LGN neurons using a spike-latency code, and how these responses can confine edge detection to borders between mid-sized objects. We propose that spike-latency-based homogeneity detection is a general neural principle that may apply to any sensory feature and modality. The mechanism is sensitive to exact spike timing, which makes it vulnerable to noise. Nevertheless we demonstrate how the mechanism can be made robust against realistic levels of neural cross talk in the brain. We show that biphasic synaptic events do yield very sharp post synaptic currents that maintain good separation of homogeneity responses in massively noisy environments. The same technique opens up a neural method of tuning neurons to varying degrees of homogeneity. The simple neural implementation, generality, robustness, and variable tuning make homogeneity-detection an attractive scheme of computation in spiking neural networks and in the brain.