Examining what our brains actually represent and how this allows us to survive and to solve complex visual tasks

We perceive the world with apparent ease—we somehow find our car keys on the kitchen table, we drive on congested roadways, we recognize familiar faces in a crowd, and as children we learn to recognize the differences between lions and tigers and bears. We have an unshakeable belief that we "see" the world—that the "images" in our mind are the accurate reflection of an external reality. The problem is thornier than one might at first imagine. For example, there is no color in the world around us; there are simply surfaces that reflect various wavelengths of light. Nonetheless, the perception of color is immediate and vivid. An enormous portion of the human brain is devoted to solving such problems, yet we have only a limited understanding of the mechanisms underlying vision.

Within Brown’s Brain Science Program (BSP), research into the cognitive, computational, and neural mechanisms of vision is diverse and interdisciplinary. Researchers with backgrounds in areas such as computer science, mathematics, engineering, neuroscience, and cognitive science all pursue the problems of how the brain understands our visual world and how we can build machines with similar abilities. Brown is unique in the degree to which these different disciplines work together in studying vision.

For instance, Brown neuro- and cognitive-scientists are designing experiments to understand how humans and other animals see and interact with the natural world. In one study, led by David Sheinberg, primates were trained to visually explore pictures of natural scenes in search of familiar objects (e.g., a parrot or a panther). The eye movements of the subjects were recorded along with the activity of neurons in their visual cortex. Studying the activity of these cells while the subjects engage in natural visual tasks leads to an understanding of the role the cells play in active visual processes where rapidly acquired sensory data directs visually guided action. Combining behavioral and electrophysiological methods provides insight into how neurons work together to efficiently select, process, and recognize visual objects and dynamic events in complex environments.

Brown cognitive neuroscientists are using behavioral psychophysics (measurements of performance), functional Magnetic Resonance Imaging (fMRI), and electroencephalography (EEG) to explore how the human visual system is fine-tuned by experience. How are we able to become experts who can immediately classify human faces, birds, or cars? fMRI brain imaging reveals that visual expertise may arise from a surprisingly large network of interacting brain areas within our visual system. BSP scientist Michael Tarr and his students have found that one small region of this network automatically computes the spatial configuration of object parts. Visual expertise involves making fine discriminations between similar looking objects and, hence, knowing the precise positions of individual parts is critical. Interestingly, this area is not active when unfamiliar objects are viewed, but becomes active once people are trained to the level of experts.

To enable computers to solve problems like those above we must uncover the fundamental computational processes underling visual perception. The study of computational vision at Brown attempts to understand these processes and produce machines that use cameras or other sensors to "perceive" the world around them. For example, humans can visually track and interpret the motion of a walking person. For computers, this task requires the complex and computationally challenging problem of "inferring" the human motion from a sequence of video images. To solve the problem the computer must exploit powerful models and algorithms to take measurements of light from the video camera and make the leap of perception to "understand" the scene. A team in Computer Science led by Michael Black has made advances in the use of probabilistic and statistical techniques that are making inference tasks such as this feasible. Nevertheless, computers are still far from achieving the effortless recognition of human vision. The interaction of machine and biological vision scientists in the BSP provides a unique environment for faculty and students to build intelligent machine vision and explain how we see.


Related Links:

http://www.vision.brown.edu/
http://www.cs.brown.edu/people/black/
http://www.cog.brown.edu/~tarr/
http://charlotte.neuro.brown.edu/
http://www.dam.brown.edu/ptg/
http://www.lems.brown.edu/vision/




Posted 11/03