Melin provides a detailed description of a set of techniques based on both type-1 and type-2 fuzzy inference systems, with the focus application of single and multi-mode biometric authentication.

Part 1 introduces concepts and techniques associated with type-2 fuzzy systems, and the combination of such systems with modular neural networks for response integration purposes. Chapters 1 through 3 provide a brief overview of fuzzy logic, fuzzy systems, neural networks, genetic algorithms, and intelligent computing in general. Type-1 and type-2 fuzzy inference systems for edge detection within images are elaborated upon, and the advantage of type-2 fuzzy logic for training data and response integration in modular neural networks is outlined. A novel approach is described in chapter 4 for response integration in modular neural networks using type-2 fuzzy logic, for biometric-based authentication including face-based, voice-based, and fingerprint-based recognition systems.

Part 2 (chapters 5 through 8) discusses the application of modular neural networks to pattern recognition. In chapter 5, the author discusses the use of a modular neural network to identify a human based on iris data. Chapter 6 provides a detailed outline on the use of a diverse set of biometric traits for human recognition. This is followed by a discussion of the significance of having an ear-based biometric system in place. The subjects are divided into three separate modules, and the outcomes of detection from each of these modules are combined using gating.

In chapter 7, the author presents a hybrid approach for combining modular neural networks with fuzzy logic for response integration when human signatures are to be verified. Modular neural networks with a set of three module outputs are combined through the use of the fuzzy Sugeno integral. Chapter 8 addresses the use of interval-based type-2 fuzzy logic for image recognition. Three monolithic feed-forward neural networks were individually trained using a supervised method. The fuzzy densities are estimated using type-1 fuzzy inference systems. Subsequently, type-2 fuzzy systems are also tested on the images. Information fusion is yet again accomplished through the use of the Sugeno integral.

The last part of the book, “Optimization of Modular Neural Networks for Pattern Recognition,” consists of chapters 9 through 13. In chapter 9, the author describes a technique based on the hierarchical genetic algorithm for “optimizing the membership functions, fuzzy rules, type of model (Mamdani or Sugeno), [and] type of fuzzy logic (type 1 or type 2),” for three biometric traits, namely, face, voice, and fingerprint. The fuzzy integrator is applied to the outcomes of the three modules of the proposed architecture to generate an outcome for the person recognition application. The fuzzy integrator of the outcomes was optimized through the use of a hierarchical genetic algorithm, and the improvements in the outcomes are enumerated.

In chapter 10, the biometric traits used include iris, ear, and voice. Three modules are defined, one for each of the three traits. The neural networks within each of these modules were trained, and the outcomes from testing were integrated through the use of fuzzy logic. Genetic algorithms were used to optimize the number of neurons in the two hidden layers. The learning algorithm type and the error targets were manipulated through the use of genetic algorithms.

Chapter 11 describes “an optimization method for the membership functions of type-2 fuzzy systems based on the level of uncertainty,” with a three-trait biometric system used as the test application. Chapter 12 addresses the effect of optimization of the neural networks on iris-based human recognition. The genetic algorithm is used for optimizing the neural network structures to improve performance.

Finally, in chapter 13, a human ear-based biometric technique is defined, wherein the fuzzy integrators used for combining the outcomes of several parallel executing modules were optimized through the use of genetic algorithms.

The language and text are easy to understand and comprehend, even for a person with little or no background in the biometric authentication field. However, a basic understanding of artificial neural networks is necessary to gain full benefit from this book.