Visual Cognition

Computational, Experimental and Neuropsychological Perspectives

Glyn W. Humphreys, Vicki Bruce

• Publisher: Psychology Press
• Year: 2021
• eBook ISBN: 9781317716624
• Topic: Psychology
• Subtopic: Cognitive Psychology & Cognition
• Index: Psychology

Vision allows us to do many things. It enables us to perceive a world composed of meaningful objects and events. It enables us to track those events as they take place in front of our eyes. It enables us to read. It provides accurate spatial information for actions such as reaching for or avoiding objects. It provides colour and texture that can help us to separate objects from their background, and so forth. This book is concerned with understanding the processes that allow us to carry out these various visually driven behaviours. In the past ten years our understanding of visual processing has undergone a rapid change, primarily fostered by the convergence of computational, experimental and neuropsychological work on the topic. Visual Cognition provides the first major attempt to cover all aspects of this work within a single text. It provides a summary of research on visual information processing, relevant to advanced undergraduates, postgraduates and research workers. It covers: seeing static forms, object recognition, dynamic vision (motion perception and visual masking), visual attention, visual memory, visual aspects of reading. For each topic, the relevant computational, experimental and neuropsychological work is integrated to provide a broader coverage than that of other texts.

1	An Introduction to Methods for Studying Visual Cognition

DOI: 10.4324/9781315785141-1

OVERVIEW OF THE BOOK

Vision allows us to do many things. It enables us to perceive a world composed of meaningful objects and events. It enables us to track those events as they take place in front of our eyes. It enables us to read. It provides accurate spatial information for actions such as reaching for, or avoiding objects. It provides colour and texture that can help to camouflage objects against their background, and so forth. This book is concerned with understanding the processes that allow us to carry out these various visually driven behaviours.

The study of visual processing has been undertaken by psychologists for over a century. During this time a wide variety of approaches has been adopted. Psychophysical and other experimental studies have examined how perception is affected by the systematic manipulation of stimulus variables, and have provided important information about the processes of normal human vision. However, for a few unfortunate recipients of brain injury, the normal processes of vision can break down. Objects or faces may appear to lack form, or to lose their familiarity. Sometimes the world may subsequently appear in shades of black, white, and grey, whereas colours were previously seen quite normally. Neuropsychological studies of such individuals provide important converging evidence concerning the identity of various visual processes; for instance, by examining which processes survive when others are impaired we can learn about the relations between different visual processes. Recently, these psychophysical and neuropsychological studies have been supplemented to some degree by neurophysiological investigations of single cells in the visual pathway of animals, and of how these cells respond to different images. More recently still, the study of visual perception has been enhanced by the work of cognitive scientists working on computer vision. Theories of vision may be explicitly formulated and rigorously tested using computer models, and even computer vision systems which make no attempt to model human vision, may sometimes yield insights which are important to us (for example, by demonstrating that a particular problem is soluble by certain sorts of methods).

This book provides perhaps the first attempt to integrate research from these different traditions within a single textbook. Indeed, as the late David Marr (1982) has argued forcefully, each kind of research has a particular role to play in the development of theories of vision. Computer simulation helps us determine whether a particular algorithm (a particular formal procedure) delivers the desired results. Psychophysical studies, of both normal and brain-injured subjects, can tell us whether the algorithm is that used by the human visual system or not. Neurophysiological studies can elucidate the way in which a particular algorithm may be implemented in the brain. Marr also emphasised, however, that an important level of theory involves understanding the abstract nature of the task faced by the visual system, and this level of understanding, which Marr termed “computational theory”, requires analysis of the optical information presented to the visual system and the mathematical methods which could, in principle, be used to unravel this information. This study of vision at an abstract or “in principle” level will also feature in our book, particularly in Chapter 2, which deals with the area in which Marr himself made such a contribution, namely the way in which the visual system analyses stimulus dimensions.

We divide the book according to the different processing functions that vision provides. Chapter 2 begins by considering relatively “early” stages of visual processing which furnish descriptions from two-dimensional images of the three-dimensional layout of the visual world from the observer’s point of view. Other processes operate on the information specified in these first stages. Chapter 3 concerns object recognition, and in particular how the viewer-centred representations of the world are transformed so that we can recognise the identities of familiar objects encountered on different occasions and seen from different points of view. Furthermore, the images of the world that fall on our retinas are constantly changing as we move relative to objects, or as objects move within a scene. We need to understand how vision enables us to see not a set of swirling, drifting, coloured patches, but people, animals, cars, etc. moving in a stable environment of surfaces and anchored objects. That is, we need to understand how vision operates dynamically. Chapter 4 is devoted to this topic. Chapters 5 to 7 are concerned with some of the other functions provided by vision. In Chapter 5 we consider the alerting function of vision, which allows us to orient our eyes and bodies in order to examine things which may prove important or threatening. In Chapter 6 we discuss the nature of visual memory and its relation to the processes involved when we perceive objects. In the final chapter, we discuss how human visual processes have been adapted to allow linguistic materials to be transcribed and read, and how the representations used for reading relate to those used in evolutionarily older activities, such as object recognition. In each chapter we try to integrate psychophysical, neuropsychological, and computational research, to provide a rich picture of the processes that make up visual cognition.

CHARACTERISTICS OF THE DIFFERENT METHODOLOGIES

Before we embark on our discussion of the visual analysis of stimulus dimensions, it is useful to contrast the emphases of the different methodological approaches we will cover, as this will provide us with a clearer idea of the goals of each approach and the way in which the approaches fit together.

Computational Studies

Attempts to implement object recognition in computers have particularly emphasised the importance of understanding real-world constraints on visual processing because it makes little sense to build a computer vision system without first understanding the nature of the information in the image that must be analysed and interpreted. Thus, a first step in computer vision requires that the “computational constraints” of the visual world are specified. The physical world has certain natural properties that are unlikely to occur accidently (e.g. when there is no object in the scene); for instance, the surfaces of objects tend to be relatively smooth, and abrupt changes in surface texture tend to occur only at the edges of the objects; also parallel or symmetrical edges tend only to occur when they are part of the same object. Because of these properties, certain “non-accidental” aspects of an image can (generally) be taken to indicate the presence of particular stimulus properties. Thus, the presence of abrupt changes in surface texture indicates an edge, and the presence of parallel edges most likely indicates the sides of an object. Indeed, by taking account of such properties of the world, problems such as finding the points of correspondence between the images in our two eyes can be solved formally (see Chapter 2).

Having formally identified these non-accidental properties, the investigator may then set about building a system capable of detecting them. The important point is that, because constraints are determined by the world and not by the particular visual system concerned (be it a computer or a person), then specifying the constraints is important for understanding human as well as computer visual systems. Many approaches to computer vision also emphasise the importance of modular processes. When writing computer programs it is often useful to divide a large program up into a number of sub-routines, which can be amended without altering the rest of the program. These sub-routines can be thought of as “modules”, each specialised for handling a particular part of a task—each accepting particular inputs and delivering particular outputs to other modules in the system. The behaviour of the system can then be characterised by the pattern of interaction between the different processing modules. This idea of modularity provides us with an initial framework for thinking about vision. For instance, different modules may exist for processing different stimulus characteristics, such as the colour, or motion of a stimulus. Indeed, a case can even be made for the existence of modules specialised for the recognition of particular classes of stimuli, such as faces or words relative to other types of objects (see Chapters 3 and 7). We can thus divide up the problems of vision according to the nature of the different stimulus characteristics that must be analysed. Even if strict modularity does not hold within the human visual system (e.g. if systems interact to affect each other’s outputs), the assumption of modular processes provides a good starting point for understanding the problem.

One other point to note is that work on computer vision differs according to how closely the solutions adopted match those found in human vision. In some cases, direct analogies can be made between the algorithms adopted in computer and human vision—David Marr’s (1982) work being a case in point (see Chapter 2). Other solutions may differ quite radically. For instance, in Chapter 3 we discuss WISARD, a program developed for object and face recognition by Wilkie, Aleksander, and Stonham at Imperial College, London (e.g. Aleksander, 1983). WISARD essentially recognises stimuli by analysing the statistical properties of two-dimensional images—a solution to object and face recognition that is probably quite unlike that found in human vision. Nevertheless, by providing an explicit formulation of how face recognition might take place, programs such as WISARD give us a model against which we can contrast and thus learn about human vision.

Neuropsychological Studies

Neuropsychological studies of vision typically attempt to understand the visual impairments caused by lesions to specific areas of the brain. One of the basic tenets of neuropsychological research is that dissociations between processes are more informative than associated deficits. It is true that patients can suffer a range of associated deficits following brain damage; for instance, disorders of cortical colour perception typically co-occur with problems in face and object recognition (see Meadows, 1974). Indeed, such associations may even be more common than patterns of dissociation. However, associations could reflect the anatomical proximity of the processes, not their functional dependence. Brain lesions do not naturally respect boundaries between the functional components of information processing, so that it is likely that more than one process may be compromised after any given lesion. On the other hand, dissociations, and in particular double dissociations (see later), can be used to argue that two processes are functionally independent of one another.

Now, when a lesion produces a loss of one ability (a single dissociation), it could simply be that that ability was particularly difficult prior to the lesion, and so could be the first to be impaired after brain insult. This argument cannot be so readily applied when a double dissociation occurs. The term double dissociation refers to a situation where one patient (patient A) is impaired at an ability that can be shown to be intact in a second patient (patient B), even though patient A is intact on tasks where patient B is impaired. An example would be where one patient has impaired colour perception along with intact movement perception, whilst another patient has impaired perception of movement along with intact colour perception. If task difficulty was the sole cause of the problems experienced by the patients, it should not be possible for the task found difficult by one patient to be easy for the other patient, and vice versa. Rather, such double dissociations suggest that the abilities are based on separable processes so that one process can be impaired without drastically affecting the other. This argument can be refined even further. It may be that two processes normally interact, but that either could be impaired without producing equivalent impairment in the other. Two such processes could be said to be functionally separable (see Patterson & Morton, 1985). However, if the spared process can be shown to operate quite normally, it can be argued even more strongly that the two processes are not simply separable, but that they operate independently in normality—as the loss of one apparently has no effect on the other. That is, the two processes operate as separable modules. An interesting question is then whether the “modules” identified by neuropsychological studies mesh with those proposed by workers in other fields. We return to this question at various points throughout the book.

Neuropsychological studies can emphasise either the nature of the impaired processes or the location of the damage suffered by a particular patient. By identifying the precise location of the damage, investigators hope to learn about the role of specific neural mechanisms in vision. For instance, the occipital lobes at the back of the brain receive the major projections from the retina, and are conventionally thought of as the primary centre for visual processing in the brain. After initially passing to the occipital lobes, visual information is then passed forward in the brain along separate pathways to the parietal and temporal lobes (respectively at the top and side of the brain). Now, the kinds of visual processing problems experienced by patients with occipital lesions tend to differ from those experienced by patients with lesions to the parietal or temporal lobes. Crudely put, occipital lesions tend to produce deficits in “early” stages of visual processing concerned with the analysis of specific visual dimensions (such as colour or depth; see Chapter 2). Parietal lesions can affect the ability of patients to orient to visual stimuli (see Chapter 5). Temporal lesions can impair processes concerned with object recognition and naming (Chapter 3). Accordingly, one might argue that the occipital cortex is the site of early visual processing, the parietal cortex the site of visual orienting, and the temporal cortex the site of object recognition.

However, there are also problems connected with localisation arguments. One is that patients with the same functional deficits (in terms of the visual processes that are impaired) may often have lesions affecting different brain sites. In such instances it is difficult to know whether there are simply individual differences in the neural implementation of visual processes, or whether the processing function is itself distributed across a number of brain sites and so can be affected by lesions to more than one area.

The second problem connected with localisation arguments is that lesions may sometimes “disconnect” processes, rather than impairing them per se. In this case, it is misleading to conclude that a particular visual process is located at the lesion site, as the lesion affects the interaction between different areas in the brain. We consider this argument most explicitly when we deal with visual attention in Chapter 5; nevertheless, the argument could be raised in other places.

Because of the above problems, and because a focus on the “neuro” of neuropsychology would change our emphasis from the processes involved in visual cognition, we do not dwell on the locus or nature of the lesions suffered by different patients.

Neurophysiological Studies

Neurophysiological studies, by their very nature, concern the way in which visual processes are implemented in the brain. Because of their emphasis on neural implementation, rather than on the nature of the processes alone, our coverage of neurophysiological work is less extensive than our coverage of the other relevant areas of empirical work. Nevertheless, much of the early psychophysical work on the analysis of stimulus dimensions was strongly influenced by investigations of the properties of cells in the retina and the occipital cortex (see Chapter 2). More recently, neurophysiological work emphasising the selective responses of cells to particular stimulus properties, such as colour or motion, has influenced thinking about the modular organisation of visual processing—linking closely to studies of patients showing selective losses of the ability to perceive these different characteristics. In such instances, neurophysiological work is directly relevant to our present concern with the functional organisation of vision.

Psychophysical and Other Experimental Studies

Psychophysical studies involve detailed investigation of how different stimulus characteristics affect performance (often that o...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Dedication
• Table of Contents
• Preface
• Acknowledgements
• 1. An Introduction to Methods for Studying Visual Cognition
• 2. Seeing Static Forms
• 3. Visual Object Recognition
• 4. Dynamic Aspects of Vision
• 5. Visual Attention
• 6. Visual Memory and Imagery
• 7. Visual Processes in Reading
• References
• Author Index
• Subject Index
• Figure Acknowledgements