Visual Categorization: Physiological Mechanisms

Recognition and Categorization

In our daily interactions with the world, we are faced with a constant stream of incoming sensory stimuli, such as faces, vehicles, furniture, and household objects. In order to plan successful behaviors that move us closer to achieving our goals, the brain is faced with the challenge of making sense of this dizzying array of objects and events around us. Because the brain typically solves this problem with remarkable accuracy and efficiency, allowing us to easily make sense of stimuli, we usually take this ability for granted. However, from a computational point of view, how to recognize the category, or meaning, of stimuli is an extraordinarily difficult problem to solve. In fact, computer scientists working with the most powerful modern computers have made surprisingly little progress in creating systems that can solve even basic recognition tasks, such as recognizing familiar faces or objects. Furthermore, the problem becomes even more difficult when you consider that objects can be viewed from multiple viewpoints, under different illumination conditions, and embedded in cluttered scenes. Thus, the human brain is by far the most sophisticated and successful recognition system in existence. Over the past several decades, neuroscientists have made some progress toward understanding the brain mechanisms underlying recognition and categorization, particularly in the visual system.

Consider the sheer number of unique visual stimuli that an average human observer can recognize, and consider also that each of these stimuli can be viewed from different vantage points, from different distances, and with different illumination. These manipulations produce a massive number of physically unique images for any one object, driving the total number of unique images toward the infinite. How does the brain make sense of this enormous range of images that we might encounter? One strategy might be for the brain to learn and store each possible image separately as “templates,” and to attempt to recognize incoming stimuli by searching for an exact match to one of the stored templates. However, the nearly infinite set of stimuli that we encounter would necessitate a nearly infinite storage capacity—something the brain clearly does not have. Furthermore, whereas this strategy might allow us to recognize an image that is exactly identical to one that we had seen in the past, we might be unable to recognize a stimulus if viewed from a slightly different vantage point or under different illumination conditions. Instead, the brain divides the nearly infinite set of possible images into discrete groups or categories.

...