The cellular neural network (CNN) architecture was introduced by Chua and Yang in 1988. At the core of this technology is an array of analog dynamic processors, or cells, that interact with their neighbors and the external world through weighted interconnections. From its conceptual base in 1988, this research has evolved to the point where it may well form the basis for a rebirth of analog computing, at least for image processing and visual computing. Nonetheless, the CNN computational model has heretofore been known primarily to a limited research community. This book, co-authored by one of the field’s founders and another leading researcher, purports to be the first textbook to make this new computing paradigm accessible to “undergraduate students from different backgrounds with a modest knowledge of mathematics and physics...”
The book is organized into 16 chapters. Chapter 1 provides an overview of the book’s contents, reserving the fundamental CNN definitions, notation, and terminology for chapter 2. Chapter 3 provides simple, yet motivational, examples of weight templates, and introduces techniques for analyzing and interpreting the state dynamics of a CNN array. Together, chapters 2 and 3 provide the critical understanding of how CNN arrays operate on two-dimensional image inputs. Chapter 4 is devoted to “implementations” of CNN dynamics. Various options, ranging from numerical solutions, to digital simulation, to analog VLSI implementations, are discussed.
Chapters 5 and 6 explore the theoretical power of the simple class called “uncoupled” networks, in terms of Boolean functions. Chapters 7, 8, and 9 serve to motivate and develop the architecture of the CNN universal machine, and show that every Boolean function can be computed by these systems. Chapter 10 discusses the design and optimization of templates for the universal machine. Chapters 11 and 12 contain advanced examples, such as a discrete space Fourier transform, and lead to extended CNN models in chapters 13 and 14.
Chapter 15 discusses both the analog and digital implementation of the CNN Universal Machine. A “visual microprocessor” based on the universal machine is described, and its performance characteristics are compared to existing processors. The final chapter introduces ongoing research into CNN models of living visual systems. Finally, a set of motivational exercises, organized by chapter, follows the endnotes and bibliography.
This book is well organized and well written. The examples are thoroughly developed and very useful for understanding the subject. The authors clearly illustrate their mastery of the topic through their clear writing style and via the overall organization of the challenging material presented.
A vision researcher, engineer, or computer scientist seeking to understand cellular neural networks should start with this book. I do take issue, however, with the authors’ statements that “no electronic circuit knowledge is needed to und erstand the first 14 chapters of this book,” and that only modest knowledge of mathematics and physics is required. An ordinary differential equation model of the CNN is introduced on page 14, and used, by necessity, throughout the remainder of the text. Analog circuit interpretations of the CNN are used starting on page 15 of the text. Readers unfamiliar with either of these topics will likely have difficulty with the remainder of the book.