Learning stereo disparity using temporal smoothness constraints: a computational model

An unsupervised learning algorithm is presented for learning stereo disparity. A key assumption is that surface depth varies smoothly over time. This assumption is consistent with a learning rule which maximizes the long-term variance of each unit's outputs, whilst simultaneously minimizing its short-term variance. The learning rule involves a linear combination of anti-Hebbian and Hebbian weight changes, over short and long time scales, respectively. The model is demonstrated on a hyperacuity task: estimating sub-pixel stereo disparity from a temporal sequence of stereograms. The algorithm generalizes, without additional learning, to previously unseen image sequences.