TU Delft
Education Type
Education print this page print this page     
2011/2012 Electrical Engineering, Mathematics and Computer Science Master Electrical Engineering
Multimedia Compression
Responsible Instructor
Name E-mail
Prof.dr.ir. R.L. Lagendijk    R.L.Lagendijk@tudelft.nl
Name E-mail
Dr. D.M.J. Tax    D.M.J.Tax@tudelft.nl
Contact Hours / Week x/x/x/x
Education Period
Start Education
Exam Period
Course Language
Required for
Statistical signal processing, information theory, speech and audio processing, selected topics in multimedia computing.
Expected prior knowledge
Linear algebra, probability theory and stochastic processes, signal processing, signal transformations.
Course Contents
In today's communication, storage and DSP technology, signals of various kinds (speech, audio, image, video, graphics) are usually represented in digital format. Although representation formats bring many advantages, an important drawback is the huge transmission or storage capacity required for digitized analog signals. For that reason, it is important to find digital representation formats that are more efficient, either via data compression or data reduction techniques. This course concentrates on the efficient and compact representation of audio-visual information in a digital format. The objective is to understand and being able to apply the concepts of digital signal (audio, speech, images, video) compression.

To this end a number of relevant compression and reduction techniques are described. Applications are found in telecommunications, multimedia, digital TV, HDTV, as well as in signal and image processing.
Study Goals
1. Properties of audio-visual signals
PCM, subsampling, digital representations, correlation in multimedia signals, redundancy, irrelevancy, quality.
- Explain factors that make compression of multimedia signals possible
- Apply PCM and subsampling, explain artefacts
- Give definition (P)SNR quality, explain relation between SNR and bit rate
- Relate visual compression artefacts to structure of compression scheme using VcDemo

2. Measures of information and VLC coding
Shannon’s information measure, entropy, lossless source coding theorem, variable length coding (VLC), universal coding
- Give definition of (differential) entropy, list properties of entropy measure
- Calculate entropy of arbitrary source and for quantizer representation levels. Give entropy of a Gaussian source
- Design Huffman code
- Explain principle of run-length coding and 2-D VLC coding
- Explain principle of LZ coding, arithmetic coding

3. Quantization
Scalar quantization, uniform and non-uniform quantization, combination with VLC encoding, vector quantization (VQ)
- Design (non)uniform quantizer by minimizing variance of quantization error
- Compare quantizers based on RD performance
- Give RD performance of quantization of Gaussian sources
- Explain problem of bit allocation and describe solution principle
- Explain principle of VQ, draw block diagram, and explain its components

4. (Differential) Pulse Coded Modulation
PCM coding block diagram, Differential PCM coding block diagram, prediction gain, linear predictor, optimal prediction coefficients, PCM and DPCM of images
- Sketch and explain block diagrams of PCM and DPCM
- Define and calculate optimal prediction coefficients and prediction gain
- Explain role of correlation in DPCM encoding signals

5. Transform Coding
Principle and block diagram of transform coding, decorrelation of signals using transforms, decomposition on basis functions, choice of transform/basis functions, KLT, WHT, DCT, extension to images, bit allocation, quantization and VLC coding of DCT coefficients, JPEG
- Describe and validate properties of orthogonal transforms
- Relate orthogonal transforms, basis functions/images, and transform coefficients
- Give properties of KLT, WHT, and DCT transform on 1-D and 2-D signals
- Motivate PCM/DPCM coding of DCT coefficients
- Relate variance of DCT coefficients to selected quantizer
- Explain role of DCT/JPEG quantization (normalization) matrix
- Explain structure of JPEG encoder and decoder, describe the coding of an individual DCT block using JPEG

6. Subband/Wavelet Coding
Principle and block diagram of subband/wavelet coding, basic 2-band decompositions system, transfer function of 2-band decomposition for ideal and non-ideal filters, filter bank using QMF filters, multiple band decomposition, extension to images, bit allocation, quantization and VLC coding of subband coefficients, application to audio compression (mp3) and images (JPEG2000)
- Draw block diagram of subband coding scheme and describe the role of the different components
- Describe structure of the basic 2-band filter bank, describe the components mathematically using linear systems theory, and give overall transfer function
- Explain “perfect reconstruction” property
- Determine how to obtain multiple band decompositions
- Explain extension to subband coding of images
- Motivate PCM/DPCM coding of subband coefficients
- Relate variance of subband coefficients to selected quantizer

7. Motion-compensated DCT-based Video Coding
MC-DCT block diagram, motion estimation and compensation, MPEG
- Motivate the structure of the MC-DCT scheme
- Explain role of motion estimation in relation to temporal DPCM
- Explain principle and problems of motion estimation
- Explain full-search, hierarchical and recursive block-based motion estimation, and estimate their complexity

Education Method
Lectures and supervised lab sessions (exercises to be solved with the VcDemo learning tool http://ict.ewi.tudelft.nl/vcdemo)
Literature and Study Materials
(1) Khalid Sayood, Introduction to Data Compression, Morgan Kaufmann Publishers Inc, US, 2006, 012620862X
(2) Lecture slides (Blackboard)
(3) VcDemo software
VCDemo software can be downloaded from http://siplab.tudelft.nl/content/image-and-video-compression-learning-tool-vcdemo. Exercises are found on blackboard.