Rodney Shaw

Signal-to-noise optimization of medical imaging systems

Over recent decades a quiet revolution has taken place in the application of modern imaging theory to many fields of applied imaging. Nowhere has this movement been more dramatic than within the field of diagnostic medical x-ray imaging, to the extent that there is now a growing consensus around a universal imaging language for the description and comparison of the increasingly diverse range of technologies. This common language, which owes much to the basic quantum-limited approach pioneered by Rose and his contemporaries, embodies the fundamentally statistical nature of image signals and enables scientists and engineers to simultaneously develop new system designs optimized for the detection of small signals while constraining patient x-ray exposures to tolerable levels. We attempt to provide a summary of some of the more salient features of progress being made in the understanding of the signal-to-noise limitations of medical imaging systems and to place this progress within a historical context. Emphasis is placed on medical diagnostics based on x-ray imaging techniques.

Ubiquitous image processing: a novel image-enhancement facility for consumers

Image-enhancement technology has been developed from first principles whereby an unskilled user may enhance and optimize the image quality of any digital photograph to personal choice within a matter of seconds. The novel methodology, which by virtue of its simple user-interface, real-time computation, and lack of any appreciable user learning-curve, naturally lends itself to many practical imaging applications in addition to that of a stand-alone application, including digital cameras, printers and photo-kiosks, or provision as an image-processing web-service. The basic imaging philosophy and principles leading to the development of this enhancement technology are discussed.

A user-friendly digital image processing procedure: technical implementation

A huge number of consumer digital images are considered unsatisfactory for general use due to their poor image quality. Of this number a significant fraction actually possess an inherent quality, as acquired, and with relatively straightforward manipulation could be enhanced and thus become entirely satisfactory according to consumer preference. In fact around fifty percent of all digital images, including those acquired with high signal-to-noise ratio using sophisticated capture devices, are positioned at a significant distance from their optimum preferred image-quality state, as defined by the end user. An unconventional practical methodology has been developed that uses a novel image processing procedure, yet leads to a solution that can readily be applied in real-time by any unskilled consumer to obtain the preferred image-quality state for each individual digital image. The underlying scientific principles are now explored in further detail, and these are related to the practical image-quality outcomes. Examples are given of the transformations applied to specific image types in order to yield preferred image-quality, and these transformations are related directly to the selection process as operated by an unskilled user of the methodology.

Perceived image quality basis for image enhancement techniques

This paper discusses the use of basic image quality metrics to construct systematic procedures for the enhancement of digital-image sharpness, and describes in detail how these same procedures may be extended to the problem of the satisfactory rendering of images covering a wide range of brightness levels (the extended-latitude problem). It is demonstrated that in both cases the solution includes the possibility of naturally-adaptive and continuously-variable enhancement techniques which are both simple in operation and undemanding of computational resources.

Simple model for noise perception in digital hardcopy

A model is developed for the stochastic noise in digital hardcopy in terms of the basic resolution element (pixel size) and the number of distinguishable gray levels printed within this element. This is expressed in Wiener (power) spectrum terms, and is used to establish a link with a perceptual scale based on previous work concerning the perception of noise for photographic (analog) images.

A real-time error-free color-correction facility for digital consumers

It has been well known since the earliest days of color photography that color-balance in general, and facial reproduction (flesh tones) in particular, are of dominant interest to the consumer, and significant research resources have been expended in satisfying this need. The general problem is a difficult one, spanning the factors that govern perception and personal preference, the physics and chemistry of color reproduction, as well as wide field of color measurement specification, and analysis. However, with the advent of digital photography and its widespread acceptance in the consumer market, and with the possibility of a much greater degree of individual control over color reproduction, the field is taking on a new consumer-driven impetus, and the provision of user facilities for preferred color choice now constitutes an intense field of research. In addition, due to the conveniences of digital technology, the collection of large data bases and statistics relating to individual color preferences have now become a relatively straightforward operation. Using a consumer preference approach of this type, we have developed a user-friendly facility whereby unskilled consumers may manipulate the color of their personal digital images according to their preferred choice. By virtue of its ease of operation and the real-time nature of the color-correction transforms, this facility can readily be inserted anywhere a consumer interacts with a digital image, from camera, printer, or scanner, to web or photo-kiosk. Here the underlying scientific principles are explored in detail, and these are related to the practical color-preference outcomes. Examples are given of the application to the correction of images with unsatisfactory color balance, and especially to flesh tones and faces, and the nature of the consumer controls and their corresponding image transformations are explored.

End-to-end linearity considerations for photon-limited detection and display systems

End-to-end system studies in the digital acquisition of photon-limited images inevitably lead to the question of setting technical design goals in order to achieve overall linearity. But this itself poses the specific question of which key print/display variables should, in ideal circumstances, be linear with which key scene-acquisition parameters. Within the digital context, intuitive answers gleaned from traditional analog approaches to this question can vary from the confusing to the misleading, and digital attempts at direct emulation of analog systems may in fact off-set the potential advantages implicit in the newer digital technologies. The author has used a digital end-to-end signal-to-noise ratio model to consider some aspects of this linearity question, and to offer solutions based on the combination of SNR optimization appropriate during image acquisition, and the human visual-response characteristics appropriate during display. Some results and conclusions are discussed here, especially those relating to the optimum digital strategy for scene-acquisition.

Model signal-to-noise comparison for CCD arrays versus film

By use of a simple model for the DQE of a CCD imaging array, the input/output imaging characteristics can be expressed in a manner which enables absolute comparisons to be made with any other image-acquisition technology. As an example a model comparison is made here between CCD-pixel-arrays and photographic-grain-arrays.In addition to DQE, it is demonstrated that other imaging parameters can be evaluated and compared in this way, including image noise, dynamic range, and output signal-to-noise ratio, and that these parameters can be related back to the respective detector mechanisms. Thus the roles played by the absolute pixel and grain dimensions and their quantum efficiencies can be identified and compared. Results of these comparisons confirm not only the obvious advantage of a multi-level mode of quantum-detection,but also present a basis for understanding the role of the CCD level-spacing function in order to fully exploit this advantage.

DQE analysis for CCD imaging arrays

By consideration of the statistical interaction between exposure quanta and the mechanisms of image detection, the signal-to-noise limitations of a variety of image acquisition technologies are now well understood. However in spite of the growing fields of application for CCD imaging- arrays and the obvious advantages of their multi-level mode of quantum detection, only limited and largely empirical approaches have been made to quantify these advantages on an absolute basis. Here an extension is made of a previous model for noise-free sequential photon-counting to the more general case involving both count-noise and arbitrary separation functions between count levels. This allows a basic model to be developed for the DQE associated with devices which approximate to the CCD mode of operation, and conclusions to be made concerning the roles of the separation-function and count-noise in defining the departure from the ideal photon counter.