Camastra F., Vinciarelli A. Machine Learning for Audio, Image and Video Analysis. Theory and Applications

Springer, 2008. — 483 p.One of the most interesting technological phenomena in recent years is the diffusion of consumer electronic products with constantly increasing acquisition, storage and processing power. As an example, consider the evolution of digital cameras: the first models available in the market in the early nineties produced images composed of 1.6 million pixels (this is the meaning of the expression 1.6 megapixels), carried an onboard memory of 16 megabytes, and had an average cost higher than 10,000 U.S. dollars. At the time this book is being written, the best models are close to or even above 8 megapixels, have internal memories of one gigabyte and they cost around 1,000 U.S. dollars. In other words, while resolution and memory capacity have been multiplied by around five and fifty, respectively, the price has been divided by more than ten. Similar trends can be observed in all other kinds of digital devices including videocameras, cellular phones, MP3 players, personal digital assistants (PDA), etc. As a result, large amounts of digital material are being accumulated and need to be managed effectively in order to avoid the problem of information overload.
The same period has witnessed the development of the Internet as ubiquitous source of information and services. In the early stages (beginning of the nineties), the webpages were made essentially of text. The reason was twofold: on the one hand the production of digital data different from simple texts was difficult (see above); on the other hand the connections were so slow that the download of a picture rather than an audio file was a painful process. Needless to say, how different the situation is today: multimedia material (including images, audio and videos) can be not only downloaded from the web from a computer, but also through cellular phones and PDAs. As a consequence, the data must be adapted to new media with tight hardware and bandwidth constraints.
The above phenomena have led to two major challenges for the scientific community:
Data analysis: it is not possible to take profit from large amounts of data without effective approaches for accessing their content. The goal of data analysis is to extract the data content, i.e. any information that constitutes an asset for potential users.
Data processing: the data are an actual asset if they are accessible everywhere and available at any moment. This requires representing the data in a form that enables the transmission through physical networks as well as wireless channels.
This book addresses the above challenges, with a major emphasis on the analysis, and this is the main reason for reading this text. Moreover, even if the above challenges are among the hottest issues in current research, the techniques presented in this book enable one to address many other engineering problems involving complex data: automatic reading of handwritten addresses in postal plants, modeling of human actions in surveillance systems, analysis of historical documents archives, remote sensing (i.e. extraction of information from satellite images), etc. The book can thus be useful to almost any person dealing with audio, image and video data: students at the early stage of their education that need to lay the ground of their future career, PhD students and researchers who need a reference in their everyday activity, practitioners that want to keep the pace of the state-of-the-art.From Perception to Computation
Audio Acquisition, Representation and Storage
Image and Video Acquisition, Representation and Storage
Machine Learning
Machine Learning
Bayesian Theory of Decision
Clustering Methods
Foundations of Statistical Learning and Model Selection
Supervised Neural Networks and Ensemble Methods
Kernel Methods
Markovian Models for Sequential Data
Feature Extraction Methods and Manifold Learning Methods
Applications
Speech and Handwriting Recognition
Automatic Face Recognition
Video Segmentation and Keyframe Extraction
Appendices
A Statistics
B Signal Processing
C Matrix Algebra
D Mathematical Foundations of Kernel Methods