Perception is carried out in OpenCog via a specialized implementation of Greedy Stochastic Pattern Mining which looks for significant patterns in the stream of inputs coming into the system.
The patterns that it searches for are those that are simple logical combinations of predicates judged as relevant. There is a control Schema that periodically updates the list of "perception-relevant" predicates. There is also a built-in measure defining what is a "simple combination," which takes into account factors such as temporal and spatial proximity as well as logical simplicity and compactness.
Mind Ontology Links