Generalized principal component analysis (Q2356522)

Generalized principal component analysis deals with many fundamental concepts, facts, and principles in engineering, computation, statistics and mathematics. This book is an introductory graduate course for students in data science, machine learning, and signal processing. It can also be used as an advanced course for students in computer vision, estimation theory, and systems theory. The book is organized into four parts, 13 chapters, and 3 appendices. Chapter 1 gives an introduction to the basic problems, and concepts studied in the book. Part I deals with methods for modeling data with a single low-dimensional linear subspace or nonlinear submanifold. It covers Chapters 2 to 4. Chapter 2 gives a review of classical principal components analysis theory and methods for subspace estimation. Chapter 3 deals with estimation of a subspace when the data are incomplete or corrupted. Chapter 4 shows how to extend previous methods to nonlinear submanifolds. Part II covers complementary approaches for modeling data with a mixture of multiple subspaces. Chapter 5 to 8 are included in this Part. Chapter 5 studies the algebraic properties of mixture subspaces. Chapter 6 presents statistical methods for estimating mixture subspace models. The focus of Chapter 7 is to explore the nonparametric spectral clustering method for subspace clustering. Chapter 8 introduces convex optimization techniques to find affinity matrices for subspace clustering. Part III presents applications of the methods and algorithms studied in earlier chapters. It covers Chapters 9 to 13. Chapter 9 shows how to cluster image patches into multiple subspaces for building sparse representations of natural images. Chapter 10 shows how to segment natural images into multiple regions corresponding to different textures. Chapter 11 shows how to segment multiple moving objects in sequences of images. Chapter 12 studies the temporal segmentation problem. Chapter 13, titled ``Final words'', discusses a few related topics that are not explicitly covered in the book because many of them are still open and active research areas. Part IV is dedicated to the appendices. They cover relevant concepts and results in optimization, mathematical statistics, and algebraic geometry in order to make the book self-contained. The book is a good text for researchers, and advanced undergraduate and graduate students of several disciplines, among others computer vision, mathematical statistics, and engineering.