240 likes | 369 Views
Experiment on 3D mobile video by subjective tests for VQEG. Orange Labs Gilles TENIOU, FT/IMG/RD/TECH/OPERA/CVA Julien LIBOUBAN, FT/IMG/RD/TECH/OPERA/HEAT Jérôme BRIARD, FT/IMG/RD/TECH/OPERA/HEAT December, 26 th 2011. Content. part 1 Context of the study part 2 Tests description
E N D
Experiment on 3D mobile video by subjective testsfor VQEG Orange Labs Gilles TENIOU, FT/IMG/RD/TECH/OPERA/CVA Julien LIBOUBAN, FT/IMG/RD/TECH/OPERA/HEAT Jérôme BRIARD, FT/IMG/RD/TECH/OPERA/HEAT December, 26th 2011
Content part 1 Context of the study part 2 Tests description part 3 Subjective tests results part 4 Conclusion Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Part 1 Context of the study Orange Labs - Research & Development – High Speed Access Video profiles – 26th Jan. 10 interne Groupe France Télécom
Context and objectives of the study • Context • Contribution in the 3GPP : Technical specification group services and system aspects in the release 11 (Mobile stereoscopic 3D video) • Objectives • Test video quality on 3D smartphone devices • Evaluate the SAMVIQ (ITU-R BT 1788) methodology on a mobile device • Find for 3D contents on mobile : • the best frame rate • the best video resolution Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Part 2 Tests description Orange Labs - Research & Development – High Speed Access Video profiles – 26th Jan. 10 interne Groupe France Télécom
Test description (1/6) – Video Sources • Video Sequences • Mix of sequences with different characteristics in terms of movement, textures, details, transitions… • 2 points evaluated : impact of the frame rate and the resolution on 3D contents. • Need of high resolution (full HD per view) and high frame rate 3D contents. Animated movie Football Interview Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Test description (2/6) – Sources adaptation • Frame rate test preparation • Resolution fixed to the Smartphone's one : qHD (960x540) • Initial 1080i50 contents are transformed to 1080p50 using TDeint filter. • The frame rate is then divided by /2, /3, /4, /5 and /6 to obtain different frame rates without impacting smoothness and avoiding using frame interpolation. • The result frame rate are : • 25 fps • 16,6 fps • 12,5 fps • 10 fps • 8,33 fps Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Test description (3/6) – Sources adaptation • Resolution test preparation • Frame rate fixed to 25 fps • Initial full HD per view content is transform to Side by Side using lanczos3 filter. • The resolution is then downscaled (lanczos3) to 5 different levels compatible with MPEG compression. • The result resolutions are : • WQVGA (320x176) • WHVGA (480x272) • WVGA (848x480) • qHD (960x540) • 720p (1280x720) Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Test description (4/6) – Encoding • Since Smartphone's hardware doesn’t allow the playback of uncompressed content, encoding process is a mandatory step. • Encoder Solution • Software encoder : Ateme Kyrion File Encoder • Encoder common features: • Constant Bit Rate (CBR) • MP4 File format • Source encoding profile : • MPEG4 AVC/H.264 at 5Mb/s CBR • Depending on the resolution of the content, the Profile/Level change to Baseline@L3.0 to High@L4.0 Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Test description (5/6) – Subjective test • Smartphone based tests • Test software adapted to Android Mobile • Test on a Sharp Aquos Phone 3D (SH80F) : • Android 2.3.4 • qHD (960x540) screen definition • 4.3” screen size, 262 ppi • Using Cell Matrix Parallax Barrier • Realtime 2D to 3D conversion • 3D depth slider • No fixe distance between the device and the tester’s eyes Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Test description (6/6) – Subjective test • SAMVIQ (ITU-R BT 1788) method • Continuous scale (0-100) graded with following french quality items : Mauvais (Bad), Médiocre (Poor), Assez bon (Fair), Bon (Good), Excellent (Excellent) • Adapted method discriminating quality levels and providing accurate quality scores • 28 Observers • 4 to 9 rejected observers in the worst conditions according the thresholds used. See annex Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Part 3 Tests results Orange Labs - Research & Development – High Speed Access Video profiles – 26th Jan. 10 interne Groupe France Télécom
Subjective Test results (1/5) – FPS (82% of correlation) • Good quality • starts at 10fps for Cartoon and Interview • not before 16,6fps for Football • Only 19/28 of testers kept Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Subjective Test results (2/5) – FPS (75% of correlation) • Results quite similar but overall quality slightly decrease. • Cartoon and Interview notations at 8,33fps are closer • 22/28 of testers kept Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Subjective Test results (3/5) – DEF (82% of correlation) • Excellent quality • obtained from WVGA with Cartoon and Interview • only reached with the hidden reference for Football • 25/28 of testers kept Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Subjective Test results (4/5) – DEF (82% of correlation) • Very low (<60) explicit reference notations deleted : • overall quality increased • better “trusty range” at 95% • 24/28 of testers kept Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Subjective Test results (5/5) • FPS global analysis • Minimum FPS required for Cartoon and Interview is 10fps whereas Football needs at least 16,6fps • Football is more than an item of quality under the other contents until the 16,6fps profile. • Definition global analysis • Poor to fair quality with WQVGA and WHVGA profiles whatever contents used • Strong gap with WVGA profile to reach good to excellent quality. • Then other profiles (qHD and 720p) don’t improve quality that much except for football with 720p profile. Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Part 4 Conclusion Orange Labs - Research & Development – High Speed Access Video profiles – 26th Jan. 10 interne Groupe France Télécom
Conclusion – R&D recommendations • Tests conducted to assess the impact of frame rate and definition when using 3D side by side contents on 3D Smartphone. • The results of this test show that Cartoon and Interview contents can be associated in terms of frame rate and definition whereas Sport content need is own specifications. • Encoding framerate and definition for a good experience : • For Cartoon and Interview : 10fps and WVGA profile. • For Sport : 16,6fps and WVGA profile • For a better experience, an encoding profil using 16,6 fps is required. • Next Step : • 3D full quality per view • Evaluate 3D quality / encoding profile Orange Labs - Research & Development – 3D mobile video subjective tests – 28th Oct. 11
Thank you Gilles TENIOU: gilles.teniou@orange.com Julien LIBOUBAN: julien.libouban@orange.com Jérôme BRIARD: jerome.briard@orange.com
End-user perceived image quality • Main goal To be acquainted of the consumer (non-experts) opinion for a target service by evaluating the image quality displayed on their own receivers, knowing that image quality mainly depends on the compression and transmission effects. • How To carry out image quality subjective assessments, involving consumers • Procedure To express each observer (consumer) opinion by assigning a score to determine quality levels according to predefined items over a quality scale of votes
Subjective method (1) • SAMVIQ (Subjective Assessment Methodology of Video Quality) • Context • Discrimination of quality for multimedia • Information on the method • Standardized at EBU and in progress at ITU-R • Continuous quality scale • Relatively short duration of tests • Global score per image sequence • Goal • Qualification of algorithms • Qualification and comparison of codecs • Potential of image quality of a network • Panel • 20 observers
Subjective method (2) • SAMVIQ Main advantages • A few number of scenes and algorithms are required to get a discrimination of quality • Very good sensitivity • Steady and repetitive results • One score per sequence • Explicite reference image • Random access and short duration per sequence (10-15s) • SAMVIQ Main disadvantages • No information about temporal variation of quality inherent in global score results