More than meets the eye

The ease with which we move around our environment, avoiding obstacles and identifying objects, belies the complexity of the visual processing necessary to support these behaviours. A large part of our brain is devoted to analysing and interpreting visual input from our retinae. The images to be analysed are continually changing and moving on our retinae as a result of motion in the outside world and of our own body, head and eye movements. Thus understanding the detection of visual motion is central to an understanding of visual perception in humans and other animals. This knowledge can be applied to the design of intelligent robots, and any principles that can be established to underlie visual function will be of interest to workers in neuroscience trying to understand the wider aspects of brain function.

Vision science is a multidisciplinary enterprise, spanning psychology, physiology, computer science and engineering. The study of visual motion detection has developed rapidly over recent years, and this book is a timely attempt to bring together leading workers in the field to write about their particular areas of expertise. The authors put forward views and theories in the context of a balanced account of current opinion and empirical evidence. The result is a stimulating and informative book that is livelier than a textbook but more didactic than a collection of journal articles.

The book is divided into six sections, each of which discuss the theme of the section at computational, psychophysical and physiological levels. In this way the reader is introduced to the complementary roles of theoretical, behavioural and neural levels of analysis in unravelling the intricacies of vision.

The introduction gives a brief history of research into motion detection, charting progress from early psychophysical studies eliciting the percept of motion from paired flashes of light, the advent of electrophysiological recording from single nerve cells and the first computational models proposed to explain behavioural data from insects. The intertwined progress of the three strands of research are followed to the present day. It is argued that computational models should be driven by physiological and psychophysical constraints, but should in turn be able to generate predictions that can be tested experimentally. In this way new theoretical insights and novel empirical findings can be incorporated into a coherent methodological framework for progress in visual science.

Estimation of velocity locally within the visual field is examined in two sections. The initial stage of motion analysis is the detection of movement within local regions of the retinal image. The operations necessary to detect this are performed in parallel at all locations across the retina, and all subsequent motion computations are based on this locally obtained information. After a description of the essential computations and neural substrates associated with motion processing, there is a discussion of the image characteristics necessary to support the perception of motion in human and primate vision.

The fourth section proceeds beyond the local detection of motion to consider the combination and integration of motion signals over space. Here there is a trade-off to be made between combining motion signals over space to give less variable estimates of image velocity and keeping local motion signals segregated to retain spatial resolution. A related issue is that of motion transparency: the question of when local motion signals are combined to yield a percept of a single coherent motion, and when they remain unintegrated to support an interpretation of multiple motions. The range of empirical evidence bearing on these issues and the level of theoretical understanding brought to bear on them here is noticeably less than in earlier chapters, reflecting the relatively recent interest in "higher-level" visual processing.

L-3drop = /One area in which the book seems to be lacking is that of visual adaptation to motion. While there is a section dedicated to spatial interactions in motion processing, there is no corresponding one on temporal interactions, despite the fact that both authors have published in this area. This seems a significant omission, as consideration of the temporal dynamics of motion perception reveals that visual motion is analysed not in a fixed or rigid way but in a way that is heavily dependent on previous visual experience. This experience may have been obtained in the previous few seconds or minutes, in which case it is usually thought of as "adaptation", or over the course of days or weeks - "perceptual learning". The plasticity of the human visual system is arguably one of its most remarkable features, but one that is broadly overlooked here.

By the penultimate section discussion has progressed from local motion detection to the interpretation of patterns of motion over the entire visual field. Patterns of motion signals characteristic of our motion through our environment are described, along with computations that extract invariants from our visual input allowing us to avoid obstacles and judge their temporal proximity. The problem of inferring spatial structure and depth from motion information is also addressed. The final section discusses the effect of eye movements on retinal motion and the use of motion information to drive eye movements. This is an important topic that is often, somewhat surprisingly, overlooked in discussions of motion detection. While the section sits slightly uneasily in its position at the end of the book, its inclusion is welcome.

L-text = /This is an entertaining and informative book, but it is likely to need regular updating to keep its relevance current.

Colin Clifford is visiting research fellow, research school of biological sciences, Australian National University, Canberra.