Skip to main content
main-content

Über dieses Buch

Image Technology Design: A Perceptual Approach is an essential reference for both academic and professional researchers in the fields of image technology, image processing and coding, image display, and image quality. It bridges the gap between academic research on visual perception and image quality and applications of such research in the design of imaging systems.
This book has been written from the point of view of an electrical engineer interested in the display, processing and coding of images, and frequently involved in applying knowledge from visual psychophysics, experimental psychology, statistics, etc., to the design of imaging systems. It focuses on the exchange of ideas between technical disciplines in image technology design (such as image display or printer design and image processing) and visual psychophysics. This is accomplished by the consistent use of a single mathematical approach (based on linear vector spaces) throughout. Known facts from color vision, image sampling and quantization are given a new formulation and, in some instances, a new interpretation.

Inhaltsverzeichnis

Frontmatter

Chapter 1. Visual Perception and Linear System Theory

Abstract
One of the goals of this book is to illustrate how many important characteristics of both the human visual system and technical systems for the acquisition and processing of images can be described using linear system theory (LST), i.e., the theory of linear vector spaces. Such a common mathematical framework has obvious practical advantages, since it simplifies the transfer of existing insights into the visual system to the optimization of technical imaging systems. A general mathematical perspective can also lead to new insights, since known mathematical theorems may have unforeseen implications. They may for instance lead to new algorithms for image processing or image coding and/or to new models for experimentally observed mechanisms in the visual system.
Jean-Bernard Martens

Chapter 2. Linear System Theory and Vector Spaces

Abstract
In this chapter, we describe a mathematical theory that is suited for describing both signals (such as images) and the operations on signals. We only discuss the case of linear operations on signals. This obviously simplifies the mathematical treatment. Many systems of interest are however linear systems (LS) or linear front-end systems (LFS) by themselves, or can be approximated by LS or LFS when operated under restricted conditions.
Jean-Bernard Martens

Chapter 3. Color Perception and Colorimetry

Abstract
In this chapter, the experimental results on color matching, as presented in chapter 1, are interpreted using the mathematical concepts introduced in chapter 2. It is demonstrated that the radiance distribution of a light source as a function of wavelength, also referred to as the radiometric function, the spectrum or the spectral power distribution, can be decomposed into a component that is relevant for color perception, called the fundamental metamere, and an irrelevant component, called the met amer ic black or residual. The fundamental metameres comprise a three-dimensional subspace of the vector space of all possible radiometric functions. This subspace is called the fundamental color space and is derived in section 2.
Jean-Bernard Martens

Chapter 4. Color Management

Abstract
Although image communication systems such as television have been in widespread use for over half a century, the interest in the accurate reproduction of color images is much more recent. There are a number of reasons for this. First and foremost, image content is usually much more important than image quality and the available image quality is most often adequate for the purpose. For instance, the frequently resurfacing debates about the need for an improved television transmission system seem to be mostly technically driven, rather than customer-driven. Second, the fact that the reproduced image can differ substantially from the original is not necessarily objectionable, especially if this original image is not available for comparison. People are often primarily interested in a pleasing picture, even if this picture is perceived to be (slightly) unnatural. Third, until recently, it was not technically feasible to improve this situation substantially at a reasonable cost. Low-cost images are mostly photographic prints and slides, and hence produced by chemical means that are hard to control accurately. The sensors and reproducing dyes are integrated in the photographic medium, so that there are few possibilities for correcting the color reproduction by means of intermediate processing.
Jean-Bernard Martens

Chapter 5. Psychophysical Measurement and Modelling of Image Quality

Abstract
Because of the large amount of imaging equipment being produced nowadays, there is a substantial economical interest in being able to measure and predict the effect of variations in the technical parameters of such systems on the resulting quality. Especially in the case of alternative systems with similar functionality, perceived (image) quality is one of the major discriminating factors between products from the point of view of the user. In this chapter, we will concentrate on how such quality impressions can be measured with the help of human subjects. In later chapters, more specifically, chapters 11 and 12, we will focus on approaches to predict image quality. Readers that are more interested in the bottom-up approach to image technology design may decide to skip this chapter.
Jean-Bernard Martens

Chapter 6. Discrete Periodic Signals and Fourier Transformations

Abstract
In the first four chapters, we have concentrated on the wavelength description of electromagnetic radiation. It was shown that three numbers, derived by proper weighting of the electromagnetic spectrum, can carry all the information needed for human color vision. These three numbers however only specify the color for one point (or uniform patch) in the visual field. Meaningful images only arise if these signals are allowed to vary in space and time. In the following chapters, we therefore concentrate on how such spatially and temporally varying visual information can be processed by technical systems in a way that is optimized for a human observer.
Jean-Bernard Martens

Chapter 7. Image Sampling and Interpolation

Abstract
In chapter 6, it was demonstrated how signals of finite extent can be mapped into periodic signals. This mapping is mostly of interest because it simplifies the description of signals in terms of a Fourier basis. Operations on periodic signals by linear time- and/or -space-invariant systems (i.e., convolutions) can for instance be described very concisely in the Fourier domain, because the harmonic signals that constitute Fourier bases are eigenvectors of such systems. Hence, periodic signals are adequate mathematical abstractions of signals of finite extent.
Jean-Bernard Martens

Chapter 8. Spatio-Temporal Characteristics of The Human Visual System

Abstract
The harmonic signals that have been introduced in chapter 6 as the eigensignals of linear space- and/or time-invariant systems have also been used extensively to study the spatial and temporal processing performed within the human visual system. These spatiotemporal characteristics of the visual system will be discussed in some detail in this chapter. They can for instance be applied to the optimization of sampling structures, as is demonstrated in chapter 9, or to the optimization of coding algorithms [Nad00].
Jean-Bernard Martens

Chapter 9. Optimizing Sampling Structures

Abstract
In this chapter, it is discussed how the theoretical results of chapter 7 on sampling and interpolation and the knowledge about human spatial perception of chapter 8 can be combined in the design of a sampling-and-interpolation system. Such a system contains an image sensor, such as a camera, that converts analog input images into discrete images, and a display that converts discrete images into analog output images that are intended for a human observer. The observer characteristics play a role in the design, since it is the combined characteristic of the display and the observer that determines the interpolation function.
Jean-Bernard Martens

Chapter 10. Image Quantization

Abstract
The signals that are of interest in image technology were shown to have continuous variations along a number of dimensions (wavelength, space, time and amplitude). By applying insights into human color perception, it could be derived that the relevant information contained in the continuous variation in wavelength can be represented by three color signals. Subsequently, based on the limited resolution of the visual system in space and time, it was argued that adequate discrete approximations to the continuous variations in space and time can be obtained by representing these color signals on a repetitive basis. Since only images of finite spatial and temporal extent are considered, a repetitive basis with a finite number of basis signals suffices in practice. Consequently, each of the three color signals can be described by a (discrete) approximation on a finite repetitive basis.
Jean-Bernard Martens

Chapter 11. Double-Ended Instrumental Models of Image Quality

Abstract
Over the past few years, we have witnessed a growing academic and industrial interest into understanding and modelling the quality discrimination process that underlies the human judgement of images. The ability to model this process is of interest for many applications, including designing new image coding or processing systems, and monitoring the performance of existing systems. We may wonder how realistic such an enterprise is, and how far we have progressed towards the set goal over the past few years1. The two concluding chapters of this book are therefore devoted to this subject of image quality modelling. They intend to illustrate alternative approaches and to identify some of the major issues still to be resolved.
Jean-Bernard Martens

Chapter 12. Single-Ended Instrumental Models of Image Quality

Abstract
In the previous chapter, we have studied double-ended instrumental measures for image quality. The two most common characteristics of existing double-ended measures are that they apply image filtering that is inspired by known perceptual mechanisms in the visual system, and that they integrate the differences between the filtered original and processed or coded image to arrive at a distortion measure. In this integration step, all differences between the filtered images are treated as being equivalent. In this chapter, an alternative perspective on image quality modelling is proposed that is in better agreement with the experimental approach towards measuring image quality that was discussed in chapter 5. This more cognitively-based perspective is inspired by two observations, i.e.,
1
that people are able to analyze and justify their image quality judgements, and
 
2
that people are able to make such judgements in the absence of an “original” (or undistorted) image.
 
Jean-Bernard Martens

Backmatter

Weitere Informationen