Keeping Video Quality Pristine throughout the Production Process: 4:2:2 10-bit AVC/H.264 encoding Pierre Larbier ATEME (Burbank, USA & Bievres, France)
Agenda • Current States • Talk about the AVC/H.264 4:2:2 10-bit profile • Empirical comparisons against 4:2:0 8-bit & 4:2:2 8-bit • Multi-Generation Encoding • Production Cases 2
Production & Transmission Processes • Production uses 4:2:2 10-bit over SDI – Used in video editing, tape decks, file storage, etc – Video needs to be transmitted by internal system or across networks • Today’s transmission mostly uses – 4:2:0 8-bit (MPEG-2 or AVC/H.264) for consumer applications – 4:2:2 8-bit(MPEG-2) for contribution applications • 20-60 Mbps with latency from 1 sec-250ms 3
AVC/H.264 Studio Profile VS. Current Implementations • Most AVC/H.264 products are currently geared towards consumer video market – 4:2:0 8-bit and <30Mbps encode rates @ High Profile – Using this equipment will necessitate resampling & incur mismatching if used for production transmission applications • High 4:2:2 Profile – Can handle 4:2:2 processing @ 10-bit and is ideal for production transmission applications – Can handle upto 200 Mbps(HD) and 40 Mbps (SD) 4
About 4:2:0 Compression • Professional video sources are 4:2:2 • 4:2:0 compression requires resampling filters 4:2:0 4:2:0 4:2:2 → 4:2:0 4:2:0 → 4:2:2 4:2:2 4:2:2 Encoding Decoding Downscaling Upscaling SDI SDI Encoder Decoder • Progressive and Interlace chroma location schemes are different • Challenges for existing chroma resampling architecture: o Possible mismatch between the 2 resampling filters o Resampling filters quality not guaranteed o Choosing between Progressive and Interlace filters is not obvious 5
4:2:0 Chroma Artifacts • Mainly caused by a mismatch between downscale/upscale filters o Color bleeding o Chroma detail loss o Interlace/Progressive artifacts • Worsen with each generation • Difficult to control in a Production chain Detail Loss Color Smearing Interlace artifacts After 5 4:2:2 ↔ 4:2:0 conversions Source picture 6
About 4:2:2 Compression At the 1 st generation, with perfect resampling filters, 4:2:2 and 4:2:0 • overall quality is roughly the same: CrowdRun, 1080i25 49 47 45 Combined PSNR (dB) 43 41 4:2:0 4:2:2 39 37 35 33 20 40 60 80 100 120 140 160 180 200 220 Bitrate (Mbps) 4:2:2 vs 4:2:0 with perfect resampling filters • But in multi-generation and/or with mismatched sampling filters, 4:2:2 quality is much better than 4:2:0 7
Current 8-bit Compression • Professional video sources are 10-bit • 8-bit compression requires rounding before compression 8-bit 8-bit 10- bit → 8 -bit 8- bit → 10 -bit 10-bit 10-bit Encoding Decoding Downscaling Upscaling SDI SDI Encoder Decoder • Motion compensation and loop-filtering performed using 8-bit samples • Challenges for 8-bit architecture: o Scalers match and quality not guaranteed o Reduced accuracy in the encoding process reduces coding efficiency o Luma shift caused by lack of rounding control 8
10-bit compression: banding removal • Processing video with 8-bit sample depth may create banding artifacts in shallow changing light scenes: o Blue skies o Underwater scenes o Sunsets • These defects are not visible at 10-bit or more Banding artifacts 8-bit compression 10-bit compression 9
10-bit compression: increased coding efficiency • Coding efficiency is increased: less bit-rate for the same quality • Rate saving between 5% and 20% on most sources • Most of the gain is provided with 10-bit coding. With 12-bit and 14-bit, coding efficiency gain is there but much smaller Woman with a Bird Cage, 1080i30 Woman with a Bird Cage, 1080i30 0.60 50 48 PSNR Y gain relative to 8-bit (dB) 0.50 46 44 0.40 42 PSNR Y (dB) 40 8 bit 10 bit 0.30 10 bit 12 bit 38 12 bit 0.20 36 34 0.10 32 30 0.00 28 0 20 40 60 80 100 120 140 160 180 200 220 0 15 30 45 60 75 90 105 120 135 150 Bitrate (Mbps) Bitrate (Mbps) RD curve at 8,10 and 12-bit Distortion gain vs 8-bit compression 10
10-bit compression: artifacts reduction • 10-bit compression reduces 3 kind of artifacts: o Contouring o Smearing o Mosquito noise • As a side-effect, 10-bit compression also reduces random distortion of shallow textured objects (clouds for instance) Mosquito Noise Contouring 8-bit compression 10-bit compression 11
AVC H422P vs MPEG-2 422P • H.264 H422P outperforms MPEG-2 422P in Contribution applications: o 10-bit support o More than 20Mbps saved at Contribution rates! Woman with a Bird Cage, 1080i30 CrowdRun, 1080i25 46 51 50 44 49 42 48 Combined PSNR (dB) Combined PSNR (dB) 47 40 46 38 45 MPEG-2 44 MPEG-2 36 AVC/H.264 43 AVC/H.264 34 42 41 32 40 30 39 38 28 0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160 Bitrate (Mbps) Bitrate (Mbps) Example 1 Example 2 12
GOP Structure Types for Contribution • GOP structure needs to be carefully considered • Possible Contribution GOP Structures – Intra-only, AVC-I, Short GOP, Adaptive Long GOP Into Tree, 1080i25 50 • 48 Adaptive Long GOP and Short GOP give 46 better performance at same bit rate 44 Average PSNR Y 42 AVC-I 100 Intra Only 40 IbBbI Long GOP 38 • Guide: let encoder make picture type 36 34 decision wherever possible 32 0 20 40 60 80 100 120 140 160 180 200 Bitrate (Mbps) Importance of GOP structure on coding efficiency 13
Role of GOP in Multi-Generation Encoding • 3-4 Re-encodes can happen in a production transmission workflow path • Aligning Encodes at GOP Boundaries can retain quality – PSNR Drop 3dB (Aligned) Vs 5dB (non-aligned) @ 7 th GEN ParkJoy 1080i25 @ 60Mbps - Aligned GOP ParkJoy 1080i25 @ 60Mbps - Unaligned GOP 36 36 35 35 Average PSNR Y (dB) 34 Average PSNR Y (dB) 34 33 33 32 32 I I 31 31 IbBbI IbBbI 30 30 Long GOP Long GOP 29 29 28 28 27 27 1 2 3 4 5 6 7 1 2 3 4 5 6 7 Generation Generation Multiple generations with aligned GOPS Multiple generations without aligned GOPS 14
Production Uses • Remote Truck Uplinks – Transmit in AVC 422 10-bit and at lower bit rates than MPEG-2 422P • Playout to Tape Infrastructures – Easily integrate transmission back into an SDI infrastructure (international applications) • Remote Collaboration – Realtime 422 10-bit at lower bit rates • Multiple Generations Re-encodes – Typical Streams can re-encode 3-4 times before it gets to consumer, so transmission in 422 10-bit aligned GOP will preserve quality 15
Advantages • Significantly lower transmission costs • Improve video quality on existing transmission links 16
Summary Using 4:2:2 10-bit allows: • Process original source video in as close to an original form • Enables most demanding application in terms of quality & rate • Offers significant gain in quality and/or rate over existing solutions • Reduces banding, smearing, and mosquito noise • Retains quality in production processes • Integrates into an SDI infrastructure at highest allowable quality 17
Thank-You p.larbier@ateme.com
Recommend
More recommend