hide
Free keywords:
-
Abstract:
In this talk I want to suggest that motion—or more generally change-over-time—are aspects
of vision that have wide ranging implications for our understanding of brain function. An observer
from another field of study might find such a claim surprising if they were to randomly
sample a selection of recent vision research papers. While the perception of motion is a well
established domain of research—as the existence of this symposium illustrates—it is nonetheless
a specialist area, taught and studied in isolation from other topics in vision. While such
specialisation occurs for many other topics as well, motion processing has another barrier to
overcome. The problem is that vision science continues to be dominated by what I will call
the “pictorial brain” assumption. Briefly stated this assumption is that visual processing begins
at the level of some static retinal “image” and ends with representations that attempt to
capture the static, spatial structure of the input. Central to this assumption is that processing
operates on a series of “snapshots”. What is missing from such a view of visual processing
is any consideration of time. I will argue that time is not a separable dimension of vision that
can be added, post-hoc, to a collection of extracted features. Our concepts of visual features
and in particular visual representations need to be modified to capture the temporal continuity
that is so much a part of the physical environment in which we have evolved. Perhaps this
caricature of “pictorial” processing described above is very far away from the way you think
about vision. But it is worth reflecting for a moment on how you do think visual processing
proceeds. What are the basic building blocks of vision? What is a feature? What is the nature
of the input? Is it discrete or continuous? Does time, temporal continuity, change-over-time,
play any role in your basic concepts? Part of the problem here is that pictorial metaphors are so
embedded in the way we work, in the language we adopt and the tools we use that they are very
rarely questioned. There can be no denying that the vast majority of experimental paradigms
continue to present static stimuli, whether in the context of human, animal and machine observers.
As the input is bound to partly determine the output, the use of such static stimuli can
be seen as problematic. More dangerous, in my opinion, is the influence that such pictorial
metaphors have on our hypotheses and models, how they constrain our thinking about vision
and visual representations. The issues that I will raise in this talk are not new ones. Many of
these concerns have been raised more than once during the relatively brief history of vision science.
As yet, however, there has been no acceptable solution that has allowed time to be more
generally integrated into our current models of visual processing. In this talk, I will also not
be able to provide such a general solution. My goal here is rather to bring this issue back into
the spotlight and to emphasise the importance of considering both space and time in vision. I
will describe recent experiments from our group and others illustrating that the visual system
often responds quite differently to dynamic rather than static input. In some domains at least,
such differences often manifest themselves in terms of measurable performance advantages for
dynamic stimuli.
