Toward a perceptually-relevant theory of appearance
Computer graphics applications usually treat appearance in different ways, depending on whether they start from scratch, from 3D scenes or from videos. I believe that a common theory of appearance would better serve such applications: it would greatly help the creation of complex appearance patterns (e.g., highlights, gradients) in drawing applications; it would permit to connect physical descriptions (e.g., BRDFs, light fields) to appearance percepts in rendering; and it would give a rich basis (e.g., differential, statistical) for image manipulation at the compositing stage.
Such a theory should describe appearance in a perceptually plausible manner, directly identifying invariants across different lighting or viewing conditions. This requires to draw connections with Human Vision. But there is also potentially room for graphics to inspire new perceptual investigations and provide tools to run experiments efficiently.