Pattern Recognition

The aim of the lecture is to present pattern recognition issues. Particularly general classification of pattern recognition approaches together with selected pattern recognition methods and techniques will be presented. Also issues concerning data acquisition, image segmentation and pattern dimensionality reduction will be discussed. Examples of applications of the discussed methods in recognition systems (together with project realised during the course) will allow listeners to analyse practical aspects of pattern recognition.
Classes have been prepared and are conducted using the competences and skills in the field of searching and verification of information, communication, information presentation, and information distribution management.

Lecture contents
Introduction

Formulation of the pattern recognition problem. Classes and objects. Classification of pattern recognition methods. Evaluation of recognition quality. Applications of pattern recognition methods.

Data aquisition and preprocessing

Overview of data aquisition hardware. Survey of parameters of the selected devices. Lossy and lossless data compression. Image quality enhancement. Image filtering and segmentation. Image transforms to other representation domains.

Selection of the feature vector; Grouping

Overview of the feature vectors used in real applications. Dimensionality reduction of the feature vector. Formulation of the grouping problem. Methods for a grouping quality and similarity evaluation.

Direct classification

Pattern method. Selection of the group representatives. Nearest neighbourhood methods. Size reduction of the training set. Selection of the k parameter in k-NN methods.

PCA and LDA analysis

Karhun-Love Transform. Implementation of Eigenfaces to face recognition. Scatter matrieces. Comparison of PCA and LDA methods.

Bayesian classification

Bayes theorem and optimal Bayes rule. Assumptions concerning probability density. Multicategory classification. Sources of a priori probability. Kernel methods for probability density estimation.

Support Vector Machines

General concept of the method. Non-linear separable case. Analysis of kernel functions used in SVM. Computational complexity of a learning and a classification phase. Accelerating SVM. Generalization properties of a classifier. Risk boundary.

Neural networks classification

Basic neuron model. Discreet and continuous activation functions. Basic learning procedures. Gradient learning algorithm. Backpropagation algorithm. Convergence and selection of learning parameters. Feedback networks. Associative memories implemented with neural networks. Self-organizing networks used for data grouping.

Syntactic methods

Basic notions: alphabets, grammars, languages, stochastic languages. Chomksy hierarchy. Grammatic inference problem for a given language. Constructing a language parser. Graph and tree grammars.

Classification quality enhancement

Multiple classifier approach. Using genetic algorithm to compute weight coefficients for individual classifiers. Voting methods. Importance of context information. Using dictionaries. Loevenstein distance. N-grams.


Laboratory contents

1. Probability density estimation. Bayesian classification.


2. Nearest neighbour classification. Training set edition and pruning.


3. PCA and LDA feature space dimensionality reduction. Application to face recognition.


4. Neural networks: selection of the architecture, training and classification. Classifier overfitting.