Pattern Recognition
Pattern recognition is a fast growing area with applications in a widely diverse number of fields such as communications engineering, bioinformatics, data mining, content-based database retrieval, to name but a few. This new edition addresses and keeps pace with the most recent advancements in these and related areas. This new edition: a) covers Data Mining, which was not treated in the previous edition, and is integrated with existing material in the book, b) includes new results on Learning Theory and Support Vector Machines, that are at the forefront of today's research, with a lot of interest both in academia and in applications-oriented communities, c) for the first time treats audio along with image applications since in today's world the most advanced applications are treated in a unified way and d) the subject of classifier combinations is treated, since this is a hot topic currently of interest in the pattern recognition community.
* The latest results on support vector machines including v-SVM's and their geometric interpretation
* Classifier combinations including the Boosting approach
* State-of-the-art material for clustering algorithms tailored for large data sets and/or high dimensional data, as required by applications such as web-mining and bioinformatics
* Coverage of diverse applications such as image analysis, optical character recognition, channel equalization, speech recognition and audio classification
Why Read This Book
You should read this book if you need a unified, mathematically sound treatment of classification, clustering and statistical learning methods that are commonly applied to DSP problems (speech/audio classification, detection, and pattern-based signal analysis). It balances theory and practical algorithms so you will learn not just the formulas but how to apply them to real signal-processing tasks.
Who Will Benefit
Graduate students, researchers, and practicing engineers working on audio/speech recognition, detection/classification problems, or any DSP task that requires statistical learning and pattern-recognition techniques.
Level: Intermediate — Prerequisites: Probability and statistics, linear algebra, basic calculus; prior exposure to signals and systems or basic DSP concepts is helpful but not strictly required.
Key Takeaways
- Formulate and implement Bayesian classifiers and understand decision-theoretic performance tradeoffs.
- Apply and tune nonparametric and parametric density estimators and mixture models using the EM algorithm.
- Design and evaluate linear and nonlinear discriminant functions including kernel methods and SVMs.
- Perform dimensionality reduction and feature selection (e.g., PCA, Fisher criteria) for high-dimensional signal data.
- Use clustering algorithms and unsupervised learning techniques to discover structure in signal and audio datasets.
- Model temporal and sequential data with Hidden Markov Models and related approaches for speech/time-series tasks.
Topics Covered
- Introduction and overview of pattern recognition
- Probability theory and Bayesian decision theory
- Parametric methods and density estimation
- Nonparametric methods and nearest-neighbor techniques
- Linear discriminant functions and perceptrons
- Performance evaluation and error bounds
- Feature selection and dimensionality reduction (PCA, Fisher)
- Clustering and unsupervised learning methods
- Mixture models and the EM algorithm
- Neural networks and learning algorithms
- Kernel methods and Support Vector Machines
- Sequential models and Hidden Markov Models
- Recent topics: learning theory, data mining and practical applications (audio/image examples)
Languages, Platforms & Tools
How It Compares
Covers similar territory to Duda, Hart & Stork's Pattern Classification but is more up-to-date on SVMs and data-mining material; compared with Bishop's PRML, Theodoridis & Koutroumbas is broader in classical pattern-recognition topics and applications.
