Communication and image processing systems are commonly modeled using random processes and are of immense interest in electronics engineering. This short volume attempts to provide an introduction to various topics related to this subject.
Organized into 15 chapters, the material is presented in a pleasant manner with several examples and illustrations. The first seven chapters cover the basics, including random variables and processes, probability distributions, Markov systems, and the modeling of noise.
Chapter 8 considers the problem of deciding the transmitted signal based on the observation at the receiver. An application called LabVIEW is depicted that allows for the simulation of systems with a binary decision.
Independent component analysis (ICA) is the topic of chapter 9. The purpose here is to find the components that are statistically independent and non-Gaussian. Three blind source separation (BSS) algorithms are presented--FastICA, joint approximation diagonalization of eigenmatrices (JADE), and extended generalized lambda distributions (EGLD)--along with applications in analyzing gene expressions.
Chapters 10 through 12 cover image processing applications. Feature extraction and classification are the two steps here. Extraction is modeled as a problem of estimating parameters of a multiscale representation using a probability density function, and the classification stage uses a similarity measure based on a probabilistic metric. The concepts of pulse-code modulation (PCM) and differential PCM are explained in detail.
The basics related to the supervised classification algorithms nearest neighbor (NN) and k-nearest neighbor (kNN) are presented in chapter 13.
Local binary patterns (LBP), which operate on grayscale images, and support vector machines (SVM), a supervised machine learning approach, are described in chapter 14. LBP has applications in face analysis and recognition, texture classification, and shape localization. For each pixel, value assignment is based on the relative intensity of the surrounding neighborhood pixels. SVM is a machine learning approach that attempts to find an optimal hyperplane that can separate image classes with largest margin (distance between support vectors and the separating hyperplane). The kernel trick handles situations where data is not linearly separable by converting input space into a larger number of dimensions.
Convolutional neural networks (CNNs), which are used to detect and classify objects in an image, are covered in chapter 15. Raw pixels are taken at the input in order to predict class scores at the output. There can be multiple hidden layers between input and output. Each neuron in the network receives certain inputs and performs a scalar product. An example of image classification is presented here.
Overall, the book follows a tutorial style, with a good listing of various formulae and equations, without developing the theory or including proofs but instead presenting numerous solved problems and MATLAB code.