380 likes | 518 Views
> HIGHERVIEW. Team: A. Sasse J. D. McCarthy D. Miras J. Riegelsberger. Presentation to UCL Network Group: 3rd March 2004. > Sharp or smooth? Comparing the effects of quantization vs. frame rate for streamed video. J.D. McCarthy M. A. Sasse D. Miras. > motivation.
E N D
>HIGHERVIEW Team: A. Sasse J. D. McCarthy D. Miras J. Riegelsberger Presentation to UCL Network Group: 3rd March 2004
>Sharp or smooth?Comparing the effects of quantization vs. frame rate for streamed video. J.D. McCarthy M. A. Sasse D. Miras
>motivation • Existing QOS policies conflict with experimental evidence. • No previous studies manipulating frame quality in conjunction with frame rate.
>motivation • IBM QOS policy (2003) “recommends reducing DCT coefficients rather than frame rate for Sports coverage, as “the priority for smooth video is higher than the priority for frame quality” • Apteker et al. (1995) • Sport coverage relatively insensitive to reductions in frame rate.
>methodology • Continuously change video quality while users are watching. • Continuously record user’s perception. • Discover the relationship between signal quality and perceived quality.
>which measure? • Mean Opinion Score (MOS) • 8-10 second clips • single camera angle • rate quality on a 5 point Likert scale. • Limitations • Doesn’t measure continuous quality variations. • Poor measure for streamed video quality. • Doesn’t measure acceptability.
>which measure? • SSCQE • The single stimulus continuous quality evaluation (SSCQE) • using a slider to indicate quality continuously. • Limitations • Too demanding for users performing real tasks. • Doesn’t measure service acceptability.
>acceptability? • Is a MOS of 3.5 acceptable to users? • What about an SSCQE rating of 70? • Service dependent?
>our approach • Focus on a specific service. • Ask users to say when the service is acceptable / unacceptable. • Advantages • Can be used with continuous streams • Easy for users to understand • Less disruptive • Relevant to service providers
>methodology • Continuously change video quality while users are watching. • Continuously record user’s perception. • Discover the relationship between signal quality and perceived quality.
>“method of limits” acceptable unacceptable high quality low quality
>“method of limits” acceptable unacceptable high quality low quality
>“method of limits” acceptable m unacceptable high quality low quality
>service functions acceptable Pr (acceptable) unacceptable high quality low quality
>service functions acceptable ITU BT.500-11 Logistic Function Pr (acceptable) unacceptable high quality low quality
>service functions acceptable ? unacceptable frame rate
>service functions acceptable ? unacceptable frame quality
>two studies • Study 1 • CIF video viewed on a desktop. • Acceptability ratings. • Eye movements. • Study 2 • QCIF video viewed on an iPAQ. • Acceptability ratings. • Qualitative interviews.
>video material • Football match • Arsenal vs Man. United (2002) • 3 source clips. • [A] Match intro and opening 3 minutes of play • [B] Highlights of Manchester United chances • [C] Highlights of Arsenal chances, final whistle and Arsenal celebration.
>participants • Study 1 • 41 football fans. • 59% watched at least once a week • 88% supported a football team. • 51% supported Arsenal or Man U.
>participants • Study 2 • 37 football fans. • 65% watched at least once a week • 84% supported a football team. • 34 % supported Arsenal or Man U.
>study 1 - results quant
>study 1 - results fps + quant
>study 1 - results gaze
>study 1 - summary • Acceptability insensitive to frame rate. • Acceptability sensitive to quantization. • Critical values: • Quantisation = 8 • Frame rate = 6
>study 2 - results quant
>study 2 - results fps + quant
>bandwidth? Critical Values (Clip B)
>qualitative comments • 84%, recognising players was impossible. • 65% had problems following the ball. • 35% said close up shots fine - but long distant shots poor. • 21% said jerky movement was a problem.
>qualitative comments “I’d rather have jerky video and better quality pictures”
>study 2 - summary • Acceptability insensitive to frame rate. • Acceptability sensitive to quantization. • Critical values: • Quantisation = 4 • Frame rate = 6
>conclusions • Limitations • Network effects not factored in. • Substantive • High motion does not need high frame rate! • Important task relevant information is lost with poor frame quality.
>conclusions • Methodological • Binary acceptability rating • continuous • easy to understand • doesn’t disrupt task • “Method of limits” produces robust replicable service functions.
>Sharp or smooth?Comparing the effects of quantization vs. frame rate for streamed video. J.D. McCarthy M. A. Sasse D. Miras