Издательство Artech House, 2008, -453 pp.If you were trying to capture a photo by inserting the film back to front and by shaking the camera, justifiably, you might be considered mad. And yet, that is what the human visual system does! The human visual system is dynamic, multirate, adaptive, and bizarre from the engineering point of view. In fact, scientists have tried to develop artificial vision systems for years, and they go about the task at orthogonal directions to what Nature has done: the human visual system relies on analog rather than digital technology; it relies on organic rather than silicon sensors; it uses irregular sampling patterns to create an image rather than sensors arranged on a rectangular grid; and it works!
This book is the result of the collective work of several scientists in the past 5 years, working towards the development of technologies inspired by the human visual system. Such an endeavor is multidisciplinary and very broad in scope, requiring the expertise of neurologists, psychologists, mathematicians, physicists, computer scientists, and software as well as hardware engineers. No single group, yet alone a single person, can solve the problems associated with understanding the human visual system in a few years, and so inevitably, this book covers only some aspects of the issues involved. Generally speaking, there are three recognised perspectives to ‘‘reverse-engineering’’ the human visual system: approaching the problem from the viewpoint of the physiologist/cognitive psychologist; approaching the problem from the viewpoint of the software engineer, and approaching the problem from the hardware perspective. We organized the contents of this book to reflect these three perspectives.
The first chapter of the book presents a general introduction and gives a broad overview of the human visual system, containing pointers to the remaining chapters of the book. Part I of the book then follows, containing chapters on the physiology/ psychology of vision. Chapter 2 provides an example of a modeling approach to the retina which is dynamical, and in which individual cells of the retina are explicitly modelled via a coupled set of differential equations that describe physiological cell-level functions. This is contrasted, in Chapter 3, which is an introduction to the broader functional organisation of receptive field characteristics of cells in V1, the primary visual cortex. Finally, Chapter 4 considers psychophysical experiments designed to probe models of V1 processing through psychophysical experiments that address visual attention. These three areas are further addressed in Parts II and III, from the mathematical and software, and from the hardware engineering perspectives, respectively.
Part II begins with Chapter 5, which models the workings of V1 as a spatial frequency analyser and discusses wavelet analysis from this point of view. Starting from the observation that the sensors in the human retina are placed on an irregular sampling pattern, Chapter 6 provides the mathematical background for performing image processing with irregularly sampled data. Chapter 7 examines the relationship between super-resolution techniques and some characteristic movements of the human eye, known as tremor and microsaccades. Chapter 8 exploits the vergence of the eyes of the viewer to estimate the depth of the viewed surface, and thus compensate for its motion, with the specific application in mind of robotic assisted surgery. Finally, Chapter 9 considers motion detection algorithms inspired by the way the human brain detects motion.
Part III deals with hardware aspects of human vision-inspired technology. Starting with the development of polymer sensors that can imitate spectral response characteristics in the human retina, presented in Chapter 10, it proceeds to consider in Chapter 11 the development of hybrid chips that combine organic (polymer) sensors and analog circuitry. Chapter 12 deals with the design of models of very large scale integrated (VLSI) analog circuits, that may be used to implement classical simple cell V1 receptive fields. Chapter 13 shows how some of the algorithms discussed in Part II may be implemented in digital hardware, while Chapter 14 considers aspects of pre-attentive vision in terms of saliency and attention caused by spatial as well as temporal saliency created by motion.
In these 14 chapters, the subject is by no means exhausted and it will probably keep scientists busy for many years to come. This book is a small contribution towards the holy grail of cognitive vision: to reverse engineer the human visual system and to reproduce its functionality for robotic applications, like robotic assisted surgery and automatic driving, medical research, like drug development and impairment diagnosis, and finally the development of artificial low powered implantable retinas.The Human Visual System: An Engineering Challenge
Part I The Physiology and Psychology of Vision
Retinal Physiology and Neuronal Modeling
A Review of V1
Testing the Hypothesis That V1 Creates a Bottom-Up Saliency Map
Part II The Mathematics of Vision
V1 Wavelet Models and Visual Inference
Beyond the Representation of Images by Rectangular Grids
Reverse Engineering of Human Vision: Hyperacuity and Super-Resolution
Eye Tracking and Depth from Vergence
Motion Detection and Tracking by Mimicking Neurological Dorsal/Ventral Pathways
Part III Hardware Technologies for Vision
Organic and Inorganic Semiconductor Photoreceptors Mimicking the Human Rods and Cones
Analog Retinomorphic Circuitry to Perform Retinal and Retinal-Inspired Processing
Analog V1 Platforms
From Algorithms to Hardware Implementation
Real-Time Spatiotemporal Saliency
