Visual Imagination in Context: Retrieving a Coherent Set of Labels with Coherencer

Abstract

A cognitive model of visual imagination will produce what we call “incoherent” results when it adds to an imagined scene that comes from multiple contexts (e.g., “arrow” and “violin” with “bow”). We approach this problem by exploring the co-occurrence of labels in images. We show that adding an incremental algorithm for examining networks of co-occurrence associations to the top-n co-occurring labels with a particular query produces greater coherence than just selecting the top-n labels or randomly selecting labels.


Back to Table of Contents