Vision Science: Photons to Phenomenology

Abstract

Vision Science: Photons to Phenomenology Steven E. Palmer. Cambridge, MA: The MIT Press, 1999. Pages: 810. Price: $70.00. ISBN 0-262-16183-4 This book is an extremely well written, and eminently readable, soup-to-nuts textbook on how we see. The text is enhanced by liberal use of illustrations that are among the best I have seen in any textbook on vision. The material in this book is drawn from a wide range of disciplines in vision science—philosophy, psychology, computer science, linguistics, and the neurosciences—and the resulting treatment of visual perception and cognition is remarkably broad in scope. Because of the author’s excellent exposition, the book functions well as an introductory text, although some of the material that he covers is quite sophisticated.FIGUREFigurePalmer’s goal is the examination of vision from three different theoretical perspectives: inferential (advocated by the 19th-century physicist and physiologist Herman von Helmholtz), ecological (advocated by the 20th-century psychologist James J. Gibson), and computational (advocated by the 20th-century physiologist and computer scientist David Marr). Although Palmer promises to be even-handed in his assessment of these perspectives, he seems to have stacked the deck slightly in favor of the inferential perspective. According to the inferential perspective, visual scene analysis is an underconstrained problem: there is not enough information in the visual data received through the eyes for perception to occur. For example, the spectral composition of the light reaching the eye from an apple depends only partly on the spectral reflectance of the apple’s surface. It also depends greatly on the illuminant, which may vary considerably even under natural conditions. Therefore, the observer needs more information if the apple is to continue to look “red” under most natural illuminants, a phenomenon known as “color constancy.” Helmholtz argued that this additional information is obtained by logical “unconscious inference,” based upon our knowledge of the properties of apples acquired from prior experience. Almost everyone who studies vision would agree that perception of the visual world cannot proceed without some underlying assumptions. However, there is currently much debate regarding the form of these assumptions and the level(s) at which they operate within the visual process of scene analysis. Adherents to the ecological view argue that the sensory data already contain much of this additional information. This information is in the form of invariant properties in the spatiotemporal patterns of stimulation that occur as the observer interacts with his environment. According to Gibson, these invariant properties can be detected directly, without the recourse to high level rule-based operations. Much of the present work on computational vision is explicitly directed toward identifying and understanding how the human visual system extracts these invariants from sensory data and exploits them in the perceptual process. Palmer clearly acknowledges this point. However, the sophisticated reader may feel that although knowledge and inferential processes undoubtedly play important roles in visual cognition, Palmer is sometimes too quick to invoke rule-based processes to explain the occurrence of certain visual phenomena, when lower-level computational strategies may work equally well or even better. The book is divided into three sections. In the first section, Foundations, Palmer defines the domain of visual perception and lays the groundwork for the discussions of theory and empirical findings that form the backbone of the book. In the concluding chapter of the Foundations section, Palmer uses color vision as the exemplar for his interdisciplinary treatment of visual perception. The second section of the book, Spatial Vision, concerns seeing things in the world based on processes that recover the image-, surface-, object-, and category-based properties of those things. This section focuses on many of the fundamental issues in classical visual space and form perception. These include image segmentation, surface reconstruction, depth perception, the perception of form and two- and three-dimensional shape, and the spatial visual constancies. A weakness in this section of the book is that it all but ignores much of the modern computational literature on three-dimensional shape and the factors that influence its perception. The influential theoretical work in this area by Jan Koenderink and Andrea van Doorn is mentioned but not developed in any depth. The final section, Visual Dynamics, is concerned with motion perception, eye movements and attention, interactions between vision and memory, and visual awareness. The treatment of many topics is delightfully interdisciplinary. For example, like most other textbook authors, Palmer begins his treatment of color vision with a discussion of traditional psychophysical and neurobiological topics: trichromacy, color-opponency, color and lightness contrast effects, hue and lightness constancy, and so forth. But, unlike most authors, Palmer then plunges into an extensive discussion of the work of Berlin & Kay and Rosch in color categorization, the Sapir-Whorf hypotheses of linguistic relativity, and determinism, and even, later in the book, an engaging discussion of that old philosophical chestnut, the “reversed spectrum” problem. I thoroughly enjoyed this aspect of the book. Vision Science: Photons to Phenomenology is a huge book, in both size (over 760 very large pages) and scope. The highly motivated student will be rewarded for his/her efforts in reading this volume with an extraordinarily broad introduction to vision science. However, there are two “down sides” to Palmer’s approach to writing this book. First, to reach a broad audience, he included almost no mathematics. This severely limits the impact of much of the modern formal work in computational vision. Second, because he wrote about visual perception on so broad a scale, Palmer had to limit his development of any given topic. Sophisticated readers are likely to find his treatments of some individual topics to be too incomplete or imprecise to be satisfying. Nonetheless, Palmer has written a remarkable book, and it will certainly have an important impact on the new generation of vision scientists.