Course description
- Introduction to pattern recognition: basic concepts and terminology, pattern representation, computational complexity of pattern-recognition algorithms, the main types of pattern-recognition methods.
- Pattern segmentation: speech-signal segmentation techniques and image segmentation techniques
- Heuristic features of patterns: features of speech segments, features of image segments.
- Application domain analysis using clustering techniques: definition of clusters and clustering, pattern-similarity measures, pre-processing of sets of patterns, hierarchical clustering algorithm.
- Optimal feature generation: class-separation measure, feature selection and feature extraction, feature generation using orthogonal transformations.
- Pattern classification by pattern matching: pattern template matching, k-nearest-neighbour rule.
- Decision-based pattern classification: decision functions, designs of pattern classifiers, polynomial decision functions, training algorithms, support vector machines, probabilistic decision functions, learning probabilistic decision functions.
- Pattern classification by neural networks: neural network topologies, back-propagation training, deep neural networks, recurrent neural networks.
- Testing pattern-recognition systems: methods for estimating the probability of the classification error with and without a test set.
Course is carried out on study programme
2nd Cycle Postgraduate Study Programme in Electrical Engineering
Objectives and competences
The objective of the course is to provide the student with the knowledge of the basic mathematical and computer concepts that are used in the construction of artificial perception systems and are essential components of intelligent systems in automation. The acquired knowledge forms the basis for understanding and designing automatic pattern recognition systems as well as the artificial intelligent systems that are based on automatic learning and knowledge acquisition from different environmental sensor data.
Learning and teaching methods
The lectures provide a theoretical background of all the considered models and methods together with simple computational examples that illustrate the key characteristics of all the presented methods. A textbook and other study material, such as lecture notes with solved example problems and lecture slides, are available to the students. As part of the lectures, the students receive optional homework assignments including theoretical questions as well as computational exercises that enable the students to promptly verify the acquired knowledge. Practical work is carried out as part of the laboratory exercises, where the students solve given programming problems. As part of the laboratory exercises, the students also carry out additional elective projects within which the selected method of automatic pattern recognition in the selected field of application should be implemented. The results of the elective projects are reported in written reports.
Intended learning outcomes
After successful completion of the course, students should be able to:
- describe the basic concepts and components of the automatic pattern recognition systems that describe their environments through the analysis of sensor data and their conversion into symbolic representations,
- explain the basic methods of feature extraction from the sensor data as well as the basic methods of pattern clustering, pattern classification and pattern recognition,
- use development tools, program environments, and databases to develop systems for automatic pattern recognition,
- analyse sensor data and other basic measurements in order to extract features that are most appropriate for the given application area,
- develop automatic pattern recognition systems for the selected application area, and
- to evaluate the accuracy and reliability of the given automatic pattern recognition systems.
Reference nosilca
- KRIŽAJ, Janez, DOBRIŠEK, Simon, ŠTRUC, Vitomir. Making the most of single sensor information : a novel fusion approach for 3D face recognition using region covariance descriptors and Gaussian mixture models. Sensors, ISSN 1424-8220, Mar.-2 2022, iss. 6, 2388, str. 1-26.
- BATAGELJ, Borut, PEER, Peter, ŠTRUC, Vitomir, DOBRIŠEK, Simon. How to correctly detect face-masks for COVID-19 from visual information?. Applied sciences, ISSN 2076-3417, Feb. 2021, vol. 11, iss. 5, str. 1-246.
- GAJŠEK, Rok, MIHELIČ, France, DOBRIŠEK, Simon. Speaker state recognition using an HMM-based feature extraction method. Computer speech & language, ISSN 0885-2308, Jan. 2013, vol. 27, no. 1, str. 135-150.
- DOBRIŠEK, Simon, GAJŠEK, Rok, MIHELIČ, France, PAVEŠIĆ, Nikola, ŠTRUC, Vitomir. Towards efficient multi-modal emotion recognition. International journal of advanced robotic systems, ISSN 1729-8814, 2013, vol. 10, no. 53, str. 1-10.
- KRIŽAJ, Janez, ŠTRUC, Vitomir, DOBRIŠEK, Simon. Towards robust 3D face verification using Gaussian mixture models. International journal of advanced robotic systems, ISSN 1729-8814, 2012, vol. 9, no. 162, str. 1-11.
Study materials
- N. Pavešić: Razpoznavanje vzorcev (3. izdaja), Založba FE in FRI, 2012.
- J. Beyerer, M. Richter, M. Nagel: Pattern Recognition, De Gruyter Oldenbourg, 2017.
- S. Theodoridis, K. Koutroumbas: Pattern Recognition (4. izdaja), Academic Press, 2009