Machine vision comprises three integrated processes: acquisition, preprocessing, and image analysis. While many resources discuss application-specific image analysis, there has been no unified account of image acquisition hardware and preprocessing--until now.
Image Acquisition and Preprocessing for Machine Vision Systems is a comprehensive, exhaustive reference text detailing every aspect of acquisition and preprocessing, from the illumination of a scene to the optics of image forming, from CCD and CMOS image capture to the transformation of the captured image.
This book bridges the gaps between hardware and software on one hand and theory and applications on the other. With its detailed coverage of imaging hardware and derivations of preprocessing kernels, it is an invaluable design reference for students, researchers, application and product engineers, and systems integrators.
Distinctive features of the book include:
• Detailed theories of CCD and CMOS image sensors, image formation, scene illumination, and camera calibration.
• Unique combination of operational details of imaging hardware (front-end electronics) and analytical theories of low-level image-processing functions.
• Coverage of image-acquisition modules and preprocessing functions within a unified framework.
• Derivation of 2D image-processing functions from first principles by extending 1D signal-processing concepts.
From an applications point of view, machine vision refers to the recovery of quantitative data from digital images. The setup for such recovery tasks requires hardware for image sensing and storage, and preprocessing software to convert captured images into image data. From an end-user's perspective, a machine vision system consists of three functionally cognate subsystems: acquisition, preprocessing, and application-specific analysis and measurement software. This book covers the first two subsystems by presenting some of the fundamental principles and characteristics of front-end hardware and derivations of a core set of preprocessing functions. Examples are included primarily to illustrate the use of some preprocessing functions rather than to provide an account of specific applications. I have taken this approach because algorithms and software for the third subsystem are application specific, and the details of many of those applications are readily available. In contrast, a unified account of image acquisition hardware and preprocessing functions is not available in any comparable literature.
In selecting the contents for this book, I excluded several areas associated with image processing, such as mathematical morphology, feature detection, shape recognition, and texture analysis, and I give only an outline description of correlation, moments, and the Hough transform. All of these topics are well covered in several other textbooks. Instead, I chose to provide in-depth coverage of the topics tied to image capture and spatial- and frequency-domain preprocessing functions for readers who are migrating to the machine vision field from other areas, as well as for practicing engineers who are seeking a conceptual account of front-end electronics and the roots of preprocessing algorithms.
The increasing degree of "smartness" of plug-and-play cameras and framegrabbers allows many preprocessing operations to be performed using default settings. However, the integration of an image-based measurement system, with emphasis on smaller memory space and shorter execution time, requires a higher level of awareness of design principles and associated performance parameters. With this context, the book covers principles related to the intrinsic characteristics of captured images, the hardware aspects of image signal generation, and the mathematical concepts of image signal processing from an algorithmic perspective of developing preprocessing software. In addition to bridging the hardware-software gap, this book provides a basis to identify some of the key design parameters and potential interface or processing limitations at an early stage of application development. In this respect, topics covered in the book are suitable for students and researchers as well as for a wide spectrum of end users, application development engineers, and system integrators from both the image processing and machine vision communities.
In building an algorithmic framework for preprocessing tasks, I adopted an approach akin to mathematical modeling and signal analysis to provide a conceptual understanding of the basic principles and their relationship to image acquisition parameters. Most of the hardware modules and preprocessing functions covered in this book are underpinned by an extensive collection of models and derivations. Other than providing insight to the design features of the front-end components (optics, sensors, and interface), this mathematical framework helps to (1) highlight some of the underlying assumptions in the operation of imaging hardware and (2) identify sources of various preprocessing parameters generally assigned by default in commercial application software. With an increasing trend toward customization and embedded design, this approach also offers a framework to select and evaluate imaging hardware and functions, and to sequence individual processing functions in the context of specific application requirements. Furthermore, since such requirements may be subsumed in the early stages of hardware design, selection, integration, and algorithm development, this book offers the theoretical foundations necessary to adapt many generic results. I hope that these design details and mathematical concepts will enable readers to effectively integrate the front-end hardware and preprocessing functions into their application platform.
Although I have included a significant number of original derivations, I have drawn much of the material from the literature. I have attempted to cite original sources as far as possible; however, due to the continued growth of the related subject areas and the large number of publications that host imaging literature, my reference lists are incomplete. While I have taken care to ensure that all derivations and supporting algorithmic descriptions are correct, some errors and omissions are likely to be present due to the involved nature of the analytical work. I will take responsibility for such errors and would appreciate it if readers brought them to my attention.
P. K. Sinha