Audio Video coding Standard of (AVS) China

Submitted by,Swaminathan Sridhar

EE 5359 Multimedia Processing Project

Video coding standards [5]

Video coding standards [4], [5]MPEG-2 (DVD, MPEG-2 (DVD, SDTV, HDTV)• More than 10 years old• Compression efficiency• 4.7GB DVD – 2 hours movie (5.3Mbps)• 18GB – 2 hours high definition movie (20Mbps)MPEG-4 AVC/H.264 (Multimedia applications)• Advanced coding techniques• Multiple-reference frame prediction• Context-based adaptive binary arithmetic coding (CABAC)• High compression efficiency• 1.5~2Mbps for SD, 6~8Mbps for HD• Save storage space, channel bandwidth, and frequency spectrum

Development stages of AVS [3]December 2003• In the 7th AVS meeting, AVS-video (part-2) and AVS-system (part-1)

was finalized.December 2004• In the 11th AVS meeting, AVS-M (part-7) was finalized March 2005• Authentication of ‘AVS101’ – high definition decoding chipMay 2005• AVS Industry Alliance was set up.June 2005• Joint AVS/ISMA workshop on IPTV standard and industry forumFebruary 2006• AVS part-2 was announced as a national standard.

Parts of AVS [3]

Applications of the commonly used parts of AVS China [3]

AVS Part-2: HD/SD video• Jizhun Profile & Zengqiang Profile• HD broadcasting• High density storage media• Video surveillances• Video on demandAVS Part-7: Mobile video• Jiben Profile• Record and local playback on mobile devices• Multimedia Message Service (MMS)• Streaming and broadcasting• Real-time video conversation

Major and Minor coding tools used in AVS part 2 [1]

Major tools• Interlace handling: Picture-level adaptive frame/field coding (PAFF)• Macroblock-level adaptive frame/field coding (MBAFF)• Intra prediction: 5 modes for luma and 4 modes for chroma• Motion compensation: 16x16/16x8/8x16/8x8 block size• Resolution of MV: 1/4-pel, 4-tap interpolation filter• Transform: 16bit-implemented 8x8 integer cosine transform• Quantization and scaling: scaling only in encoder• Entropy coding: 2D-VLC and Arithmetic Coding• In-loop deblocking filterMinor tools• Motion vector prediction• Adaptive scan

AVS encoder structure [1]

Different picture types [2]

Three types of picture are defined by AVS namely • Intra pictures (I) •Predicted pictures (P)- At most two reference frames (P or I) •Interpolated pictures (B)- two reference frames (I or P or both)

MB level Adaptive frame coding [2]MB-level adaptive frame/field coding (MBAFF)• The frame/field encoding decision is made independently for each vertical pair of

macro blocks in a frame.• A frame consisting of both moving and non-moving regions is coded more

efficiently by:• frame mode for the non-moving regions• field mode for the moving regions• MBAFF is much more complicated than PAFF– zig-zag scanning– motion vector prediction– intra prediction– deblocking– context modeling in entropy coding• The advantage compared with the MBAFF in H.264– A field-coded MB belonging to the bottom field CAN use the top field ofthe same frame as a reference for motion prediction

Intra Prediction [2]• Five different modes for luma

Luma Intra Prediction difference between AVS and H.264 [6]

AVS• Block size: 8x8• 5 modes• Reference pixels low pass filtered• Advantages: low complexity with less modesH.264• Block size: 4x4 or 16x16• 9 modes for 4x4 & 8x8, 4 modes for 16x16• Advantage: better prediction• Disadvantage: more complex

Intra prediction modes for Chroma [2]• 4 Prediction modes for Chroma

Inter Prediction and Motion Compensation [1]

• At most 2 frames can be stored as reference for motion prediction.• Block size of motion prediction and compensation– 16x16, 16x8, 8x16 and 8x8• In each MB, the number of MV pairs can be 1, 2 or 4, depending on the block size

of MC.• MVD, the difference between the predicted MV and the real MV, is coded.• Resolution of MV– 1/4-pixel for luma– 1/8-pixel for chroma• Motion prediction modes• – Forward• – Backward (only applicable for B frame)• – Bi-directional (only applicable for B frame)• Skip• Direct• Symmetric

Reference Frame [1]

• At most 2 reference frames are used. • PAFF or MBAFF is used,– if the current MB is frame-coded, 2 frames can be

used as reference for motion prediction.– if the current MB is field-coded, 4 fields can be

used.• Reference index should be coded with every MC

block to indicate which reference picture is used

Reference Index [1]

Motion Vector Prediction [3]

Motion Vector Prediction [3]• Use A, B, C, D’s MV (MVA, MVB, MVC and MVD) to predict E’s MV

(PredMVE)• Reason: reduce the bits for coding MV• Method:• Geometrical median of MVA, MVB, MVC• VAB = Dist(MVA, MVB)• VBC = Dist(MVB, MVC)• VCA = Dist(MVC, MVA)• FMV = Median(VAB, VBC, VCA) where Dist(MV1, MV2)=|x1-x2|+|y1-y2|.• Determine PredMVE• If FMV equals VAB, PredMVE=MVC.• If FMV equals VBC, PredMVE=MVA.• If FMV equals VCA, PredMVE=MVB.

Interpolation for Luma [3]

• Resolution– Quarter-pixel• Filter– Half-pixel• Blue: [-1, 5, 5, -1]– Quarter-pixel• White: [1, 7, 7, 1]• Red: bilinear

Interpolation for Luma [3]Full Full Full Full

Full

1/2

1/2 1/2

1/2 Full Full Full1/4 1/4

Full Full

1/2

1/2 1/21/2

1/2

1/41/4

1/4

1/4 1/4

1/4

1/4

1/4

1/4 1/4

Full

Full Full Full

Full

Full

1/2 1/2 1/2

1/2

Full

1/4

1/2

Full Pixel

1/2 Pixel

1/4 Pixel

Interpolation for Chroma [3]

PredMatrix[x,y]=[(8–dx)×(8–dy)·A + dx · (8-dy)·B + (8–dx) ·dy·C + dx·dy·D]/64

Forward and Backward Prediction [1]

• Forward prediction• MV pointing only to the previous frame• Get reference block only from the previous

frame• Backward prediction• MV pointing only to the next frame• Get reference block only from the next frame

Forward and Backward Prediction [1]

Bi-directional Prediction [1]• Skip mode• Block size of MC: 16x16• No transform coefficient is coded, since they all equal zeros.• No MV is coded, since they can be calculated.• Direct mode• Block size of MC: 16x16 or 8x8• Transform coefficients are not all zeros, so they have to be coded.• No MV is coded, since they can be calculated the same way for skip mode.• Symmetric mode• Block size of MC: 16x16, 16x8, 8x16, 8x8.• Transform coefficients are not all zeros, so they have to be coded.• Only forward MV is coded, and the backward MV can be calculated by

using the forward one.

MV Derivation for Skip and Direct Mode [1]

MV Derivation for Symmetric Mode [1]

Pre-scale Transform [3]

AVS 8×8 ICT Matrix [3]

Context-based Adaptive 2D VariableLength Coding (CA-2D-VLC) [1]

(level, run) pair mapping to CodeNum using VLC tables

• level>0: CodeNum is the number in VLC tables directly

• level<0: CodeNum is number+1 in VLC tables .• • Example• level= 2, run=1, CodeNum=11;• level= −2, run=1, CodeNum=12CodeNum mapping to bit• string using Exp-Colomb coding

Context-based Adaptive 2D VariableLength Coding (CA-2D-VLC) [1]

Deblocking Filter [3]8x8 block• Three steps• Choose boundary strength• (BS), according to• Prediction modes• MV• Decide whether to filter• according to• Quantization Parameter (QP)• BS• – Apply filter to the boundary

Deblocking Filter [3]

AVS Part-2 vs H.264/AVC [4], [6]

AVS Part-2 vs H.264/AVC [4], [6]

# BS: Boundary strength

AVS Part-2 performance [1], [A]

Container.qcif sequence [A]

Decoded frame [A]

Claire.qcif sequence [A]

Decoded frame [A]

News.qcif sequence [A]

Decoded frame [A]

References1] L. Yu et al, “An Overview of AVS-Video: tools, performance and complexity”, Visual Communications and Image Processing 2005, Proc. of SPIE, vol. 5960, pp.596021, July 31, 2006.2] L. Yu et al, “An area-efficient VLSI architecture for AVS intra frame encoder” Visual Communications and Image Processing 2007, Proc. of SPIE-IS & T Electronic Imaging, SPIE vol. 6508, pp. 650822, Jan. 29, 2007. 3] W. Gao et al, “AVS - The Chinese Next-Generation Video Coding Standard” NAB, Las Vegas, 2004. 4] T. Wiegand et al, “Overview of the H.264/AVC Coding Standard” IEEE Trans. Circuits Syst. Video Technol., vol.13, pp.560-576, July 2003. 5] J. Wang et al, “An AVS-to-MPEG2 Transcoding System” Proceedings of 2004 International Symposium on Intelligent Multimedia, Video and Speech Processing , Hong Kong, pp. 302-305, Oct. 20-22, 2004.6] X. Wang et al, “Performance comparison of AVS and H.264/AVC video coding standards” J. Comput. Sci. & Technol., Vol.21, No.3, pp.310-314 J, May 2006.7] B. Tang et al, “AVS Encoder Performance and Complexity Analysis Based on Mobile Video Communication”, WRI International conference on Communications and Mobile Computing, CMC ‘09, vol. 3, pp. 102-107, 6-8 Jan. 2009.

Web References:AVS China softwareA] ftp://159.226.42.57/public/avs_doc/avs_software