This project is about a kohonen network which adapts to the input it sees through the webcam. It computes the difference between two consequent frames and passes the pixels which changed to the network, which will adapt focusing on the zones that changes more.

The last canvas shows the neurons colored with the mean color of the pixels in their cluster. This could be used as input for a feedforward neural network, which would further elaborate the information coming from the zones which changed. This is of course a very limited imitation of how real vision works, even animal's vision is much more complex than that.

Project by Daniele Nicassio. more infos and projects on my blog