Biological and Computer Vision

Gabriel Kreiman

Cambridge University Press. 2021. ISBN 9781108649995 

Additional Materials

Chapter III: The phenomenology of seeing

Marr and Poggio defined three fundamental levels of analyses to understand vision: the system's behavior, the computations, and implementation at the neuronal level. The exciting field of psychophysics quantifies behavior through measurements of reaction times, performance, and eye movements. For example, Gestalt laws describe how visual features tend to be grouped to segment objects in an image. Visual recognition is fast: in approximately 200 milliseconds, observers get a glimpse of their surroundings and identify specific objects in a transformation-invariant manner. The recognition process can solve complex problems like inference from partial information due to occlusion, incorporate spatial contextual information, and integrate information over time. Although intuition might suggest that vision is a veridic testament to what is out there in the world, many visual signals are not captured by the brain, and the brain also makes up stuff that does not exist. Visual perception is a construct where brains build an interpretation of the surrounding world based on incoming information and internal rumination. A plethora of visual illusions provide clues about this interpretation process and thus constrain the mechanistic understanding of visual processing.