A model for the spatial integration and differentiation of velocity signals.

We present a model of optic flow processing which is able to reconcile the integrative, cooperative phenomena of motion capture and coherence with the differentiation of velocity signals in motion segmentation and transparency. The model uses a Markov random field to compute the behaviour of coextensive topographic neural maps of retinotopy and velocity. We have used the model to simulate the psychophysics of motion coherence, motion capture and transparency. Further, it exhibits motion segmentation without extra postulates. The model is robust and able to display all types of motion percept with the same parameter set.