CitationDownload as .RIS
Emerald Group Publishing Limited
Copyright © 2009, Emerald Group Publishing Limited
3D Shape: Its Unique Place in Visual Perception
Article Type: Book review From: Sensor Review, Volume 29, Issue 2
Zygmunt PizloThe MIT Press, Cambridge MAApril, 2008312 pp.ISBN: 978-0-262-16251-7£24.95http://mitpress.mit.edu/catalog/item/default.asp?ttype = 2&tid = 11496
Although this book has a chapter entitled “Machine Vision”, it is philosophical and psychophysical in its approach. Vision engineers will not use this as a direct source of reference. However, for background reading, and particularly as a review of the development of ideas, this book is worth reading.
Pizlo traces the history of theories and experiments in 3D shape perception, distinguishing influential work from helpful work! (“Millstones from milestones” as he puts it.) The book concentrates on the perceptual phenomenon of shape constancy – that is our perception that a particular object “out there” has a fixed shape despite the fact that it casts different shaped images on the retina when viewed from different angles. For Pizlo, the important thing about shape constancy is that is tells us something about the object itself – it allows us to perceive objects veridically.
The author elucidates the differences between the original Gestalt psychologists’ approach to perception and those of the Associationist, Cognitive Revolution and Empiricist schools. He studies the approaches of Marr, Gibson and Biederman and looks at the relative importance of binocular disparity, viewpoint, shading, texture and motion on shape perception. He goes on to describe more recent experiments using unstructured wire objects and irregular surfaces (“potato chips”).
In his final chapter, Pizlo presents a new approach to shape perception, treating it as a mathematically ill-posed inverse problem whose solution requires a priori constraints. Figure-ground organisation (segmentation) is the first step in its solution, and edge detection coupled with closure and symmetry constraints deal with this. He proposes symmetry and planarity constraints for the next step of working out the 3D shape of each figure, and adds the new constraint of compactness, calculated from the volume and surface area of the perceived object.
I did not find this to be easy reading. The complexities and uncertainties of the theories left me wondering how practical machine vision systems manage to work so well!
Dr Christine ConnollyStalactite Technologies Ltd, UK