How Vision Works.docx

This lecture explores the nature of human visual perception, and understand the challenges that need to be addressed to artificial visual systems How Vision Works Part I Summary • The “inner screen” theory of vision fails because there is no “inner eye” to inspect the inner screen • The topographic...

This lecture explores the nature of human visual perception, and understand the challenges that need to be addressed to artificial visual systems How Vision Works Part I Summary • The “inner screen” theory of vision fails because there is no “inner eye” to inspect the inner screen • The topographic maps in our brain are upside-down, flipped L- >R, and distorted • The brain extracts multiple forms of information about vision including separate streams for “what” and “where” information • As you ascend the “what” pathway neurons respond to increasingly complex features from edges to faces and places • The responses of cells in the different layers of the visual system can be usefully compared to those of artificial neurons found in deep convolution networks How Vision Works Part II Summary • Our brain uses powerful principles to build percepts including closure, similarity, proximity, and continuity • The brain tries to find an interpretation of the visual scene that is consistent with past experience and with assumptions of invariance about shape, colour, size, illumination and so on • The brain seeks a consistent interpretation of the whole visual scene, the emergence of this interpretation allows it to resolve perceptual ambiguity How Vision Works Part III Summary In Part 3 we explore how the brain uses knowledge to make hypotheses and inferences about the world that it allow it to make sense of visual experience Perception is inherently ambiguous • The brain uses knowledge about the world to make sense of the visual scene • Some of this knowledge might be hard-wired in brain, the rest is acquired from experience • Computer vision systems will need to make use of similar strategies to reduce uncertainty