Gong Y., Xu W. Machine Learning for Multimedia Content Analysis

New York: Springer, 2007. — 279 p. — ISBN: 978-0-387-69938-7, e-ISBN: 978-0-387-69942-4.The objectives we set for this book are two-fold: (1) bring together those important machine learning techniques that are particularly powerful and effective for modeling multimedia data; and (2) showcase their applications to common tasks of multimedia content analysis. Multimedia data, such as digital images, audio streams, motion video programs, etc, exhibit much richer structures than simple, isolated data items. For example, a digital image is composed of a number of pixels that collectively convey certain visual content to viewers. A TV video program consists of both audio and image streams that complementally unfold the underlying story and information. To recognize the visual content of a digital image, or to understand the underlying story of a video program, we may need to label sets of pixels or groups of image and audio frames jointly because the label of each element is strongly correlated with the labels of the neighboring elements. In machine learning field, there are certain techniques that are able to explicitly exploit the spatial, temporal structures, and to model the correlations among different elements of the target problems. In this book, we strive to provide a systematic coverage on this class of machine learning techniques in an intuitive fashion, and demonstrate their applications through various case studies. The main body of this book is composed of three parts: I. Unsupervised learning, II. Generative models, and III. Discriminative models.
This book is devoted to students and researchers who want to apply machine learning techniques to multimedia content analysis. We assume that the reader has basic knowledge in statistics, linear algebra, and calculus. We do not attempt to write a comprehensive catalog covering the entire spectrum of machine learning techniques, but rather to focus on the learning methods that are powerful and effective for modeling multimedia data. We strive to write this book in an intuitive fashion, emphasizing concepts and algorithms rather than mathematical completeness. We also provide comments and discussions on characteristics of various methods described in this book to help the reader to get insights and essences of the methods. To further increase the usability of this book, we include case studies in many chapters to demonstrate example applications of respective techniques to real multimedia problems, and to illustrate factors to be considered in real implementations.