On Considering Semantics for Multi-image Processing