Vision research

"In shorter time, more will be known about the most remote objects, namely the stars,
than about the most nearby topic, namely perception" -- Aristotle (384--322 BC)

Human vision research aims at understanding the neuro-cognitive process that takes the light in our eyes as input and that enables us to perceive scenes as structured wholes consisting of objects arranged in space. This perceptual organization process is believed to be one of the autonomous brain processes that underly consciousness and, thereby, virtually every impression we experience and virtually every action we undertake. In everyday life, we take this process for granted. We take vision as a reliable source of information about the world, even though vision is not 100% veridical (i.e., truthful). For instance, eye-witnesses often give different, if not contradictory, accounts of the same event. Furthermore, the next figure gives one of the many visual illusions showing that "what we see" is not always "what we look at". See also Kaleidoscope for a class of arresting motion and velocity illusions.

In fact, it is amazing that vision is usually sufficiently veridical to guide action. Aristotle already realized that our eyes are not windows through which we see objects as they are, but that, inversely, we take objects to be as we see them. The point is that vision does not start with objects but with millions of tiny light receptors in the retina of each eye. At best, these light receptors can be said to provide a two-dimensional image of colored patches. Yet, after "automagic" processing in our brain, we experience a world of three-dimensional objects. This automagic process, from images to objects, is the subject of vision research. The next figure illustrates the difference between vision as a tool and vision as a topic.

To study this automagic process of vision, the next three methodological distinctions (without clear-cut borders) are useful to specify the position of scientific findings concerning subquestions in the total field of vision research:

(1) The total field of vision research may be divided into three subfields (see the right-hand picture above):

• low-level vision --- which concerns the extraction of image properties from the retinal image.
• middle-level vision --- which concerns the integration of image properties into perceptual organizations.
• high-level vision --- which concerns the everyday functionality of perceptual organizations.
  

(2) Observations and phenomena may be analyzed at three levels of description:

• the computational level --- which focuses on the nature of the mental representations that result from cognitive processes.
  
• the algorithmic level --- which focuses on the processing mechanisms of these cognitive processes.
• the implementational level --- which focuses on the neural realization of representations and cognitive processes.

(3) Theories and models may be enhanced, revised, or rejected via three cycles of research:

• the theoretical cycle --- which aims to assess the conceptual plausibility of ideas and assumptions.
• the empirical cycle --- which aims to test ideas and assumptions by way of controled experiments.
• the tractability cycle --- which aims to assess if ideas and assumptions allow for feasible implementations.

For more details on methodological principles guiding my research, see Marr's levels, Research cycles, and Metaphors of cognition




Passing all three research cycles, my research focuses on middle-level vision (with low-level and high-level offshoots) and on the computational and algorithmic levels of description (with implementational offshoots). That is, the core of my research concerns ideas about representational and processing aspects of the integration of image properties into perceptual organizations.

The integration of image properties into perceptual organizations is an intriguing problem. The visual system performs this integration very rapidly and fairly veridically, even though it is faced with a fundamental ambiguity. That is, in everyday situations, the retinal image may form a rich source of information but it is nevertheless just a 2-D projection of a 3-D scene. This implies that the retinal image underdetermines the 3-D scene, yielding a fundamental ambiguity that is illustrated in the next figure.




An open question is whether the visual system indeed considers all possible interpretations of a stimulus to select a specific interpretation. Yet, in my research, vision is modeled as if it performs such a selection, to gain more insight in how the visual system solves the ambiguity problem. Then, pressing questions are, for instance:

• What is the selection criterion?
• How can the selection be performed so rapidly?
• Why does the selection criterion yield such veridical outcomes?

Such are the questions my research focuses on, by means of mathematical formalizations, computer implementations, cognitive models, and psychophysical experiments.

For more details on my specific research topics, see Research