Data Science, Compression and Image Communication
Overall Course Objectives
To give the participants solid knowledge of the basic principles of data science and compression and methods for data compression e.g. for image communication. To enable the students to implement and evaluate data analysis and compression algorithms.
Learning Objectives
- implement a data compression or analysis scheme
- simulate data compression encoding and decoding
- calculate and evaluate the performance factor
- calculate and evaluate the quality of decoded or reconstructed data objectively and subjectively
- analyse and explain a code length in relation to an entropy estimate
- explain decorrelation of data for analysis and compression
- explain quantization of a data representation
- explain the entropy concept, e.g. for coding
Course Content
Introduction to the basic principles and methods for data science and compression incl. basics of information theory as entropy, cross entropy and mutual information (MI). The entropy is a measure of information and mutual information a measure of shared information. This includes entropy coding by Huffman and arithmetic coding, decorrelation by prediction or transformation to the frequency domain and quantisation of data for lossy coding and represenation.
Some examples of data for compression and analysis are presented, eg. coding and analysis of: Data strings, Images in the frequency domain (DCT and wavelets – JPEG 2000) Audio signals, and video as image sequences. One application area is coding and analysis of images, video and data from drones.
Midway, the students choose a project within the areas presented.
Teaching Method
Lectures and individual project (in groups of two)