Perceived Quality for Transported Video

Perceived Quality for Transported Video

Nele Van den EndeReinder HaakmaMaddy JansePeter van der Stok

2Research

Overview

• Background• Adaptation Methods

– I-Frame Delay– Signal-to-Noise Ratio Scalability

• Research Categories• Perception Experiments

– Method– Design– Results

• Summary• Future Research Questions

3Research

Background

Access to content, anywhere, anytime

4Research

Background

But there are disturbance problems…

5Research

MPEG Encoding

• Compress video• Play back platforms• Types of frames

– Intra-coded, predictive-coded & bi-directionally predictive coded

• Groups of Pictures

I B B P B B P B B

6Research

MPEG Induced Effects

7Research

MPEG Induced Effects

8Research

Adaptation MethodsI-Frame Delay

• Network clogging & buffer overflow drop frames with least importance

• Two parts: tagger & dropper

IFD queue

SSSSWWWW

Outgoing packets

C

Incoming packet

SWW

9Research

Adaptation MethodsSignal-to-Noise Ratio Scalability• Dividing video in layers: one base-layer, several

possible enhancement layers

10Research

Research Categories

• Compressing methods• Adaptation methods• User perceived errors

11Research

Perception ExperimentsMethod

• Double-stimulus continuous quality-scale• 10 sec video sequences• 1 hour per observer

12Research

Perception ExperimentsDesign

• Bitrate: 6 vs. 3 Mbps• Scenes: Matrix - Reloaded vs. Feet of Flames• Duration of loss: 2 vs. 4 vs. 8 sec

• IFD: 1/7 vs. 1/14 B-frame loss• SNR: 1/3 vs. 2/3 base-layer

13Research

Perception ExperimentsQuestions - IFD

• Is leaving out more B-frames (compared to leaving out less B-frames) perceived as worse quality?

• Does the duration of a quality drop influence perceived video quality?

• Do observers notice the difference between the shown bit-rates?

14Research

Perception ExperimentsResults - IFD

15Research

Perception ExperimentsQuestions – SNR Scalability

• Does the duration of a quality drop influence perceived video quality?

• Does a quality saturation effect really show? Do observers really not notice when video quality is enhanced once a certain level is reached?

• Do observers notice the difference between the shown bit-rates?

16Research

Perception ExperimentsResults – SNR Scalability

17Research

Summary

• Saturation effect – when perceived quality is already low/high, observers don’t notice

further decreasing/increasing of the objective quality– thresholds depend on scene content

• IFD– effects were small, but B-frame loss was small – influence of duration and amount of quality loss depends on

scene content and bit-rate – B-frame loss is more pronounced and lasts longer differences

between reference and modified video sequences seem easier to detect

• SNR– observers perceive the different bit-rates of the base-layers – extent depends on scene content and duration of the quality drop

18Research

Future Research Questions

• Influence of the content’s typesemantic versus physical appearance?

predictability versus unpredictability?

• Management of the disturbances or the type of perceived disturbances?

• What about audio?