A simple visual perception model by adaptive junction

The authors construct a simple visual perception model for random image sequences of parts of objects, using the adaptive junction network. These networks are continuous-time asymmetric neural networks recognizing spatio-temporal patterns. They prove that adaptive junction networks have three kinds of internal representation and recognizes four faces in terms of spatio-temporal patterns consisting of eyes, noses and mouths. The results indicate not only that an adaptive junction network has less hardware complexity than other conventional visual models, but also that this adaptive junction network can demonstrate one kind of optical illusion