Multimedia Networking: Digital Signal Representation, Compression, and Transform Coding - , Study notes of Electrical and Electronics Engineering

Download Multimedia Networking: Digital Signal Representation, Compression, and Transform Coding - and more Study notes Electrical and Electronics Engineering in PDF only on Docsity!

Multimedia Networking

ECE 599

Prof. Thinh Nguyen

School of Electrical Engineering

and Computer Science

Based on lectures from B. Lee, B. Girod, and A. Mukherjee

Outline ‡^ Digital Signal Representation^ „^

Audio/Video

‡^ Lossy Compression ‡^ DPCM ‡^ Transform coding ‡^ JPEG

Signal representation^ ‡^ For perfect re-construction, sampling rate (Nyquist’sfrequency) needs to be twice the maximum frequency ofthe signal.^ ‡^ However, in practice, loss still occurs due to quantization.^ ‡^ Finer quantization leads to less error at the expense ofincreased number of bits to represent signals.

Audio sampling ‡^ Human hearing frequency range: 20 Hz to 20 Khz. ‡^ Voice:50Hz to 2 KHz ‡^ What is the sampling rate to avoid aliasing? (worsethan losing information)

Audio CD :

44100Hz

Pulse Code Modulation (PCM) ‡^ The 2 step process of sampling and quantization isknown as

Pulse Code Modulation. ‡^ Used in speech and CD recording.

audiosignals

sampling

quantization

bits

No compression, unlike MP

DPCM (Differential Pulse CodeModulation)

DPCM - Image

1/4 0 1/ 1/ Linearpredictor

DPCM - Image

Transmission errors in a DPCM system

DPCM-Video ‡^ Interframe coding exploits similarity of temporalsuccessive pictures. ‡^ Important interframe coding methods^ „^

Adaptive intra-interframe coding „ Conditional replenishment „ Motion-compensated prediction

Transform Coding^ ‡^ Why transform Coding?^ „

Purpose of transformation is to convert the data into a formwhere compression is easier ‡ Transformation yields energy compaction „ Facilitates reduction of irrelevant information ‡ The transform coefficients can now be quantizedaccording to their statistical properties. „ This transformation will reduce the correlation between thepixels (decorrelate X, the transform coefficients areassumed to be completely decorrelated (RedundancyReduction).

How Transform Coders Work^ ‡^

Divide the image into 2x1 blocks^ „^ Typical transforms are 8x8 or 16x

Joint Probability Distribution^ ‡^

Observe the Joint Probability Distributionor the Joint Histogram.

Correlated Pixels^ ‡^

Since adjacent pixelsx^ and x^1

are highly 2 correlated the jointprobability p(x

, x) 1 2

is concentratedaround the line FF’and variancesare equal becausethey are the samplesof the same image.