MPEG 및 H.264 압축 2006 년 11 월 동국대학교전자공학과원치선 cswon@dongguk.edu
기존의압축표준비교 Communication Storage/Entertainment Date of Standard Primary Applications Typical Video Rate Typical Video Frame Size H.261 H.263 MPEG-4 MPEG-2 MPEG-1 1990 5/94 revised 5/96 Version 1 1/98 Version 2 2/99 Version 1 2/00 Version 2 11/94 11/92 Wireline VideoDesktop/WirelessWeb Authoring Broadcast, Digital, Videophone & DVD, Wireless Conferencing DVD, HDTV Storage Media conferencing Videophone, HDTV 128-384 kbps 20-384 kbps 20 kbps-6mbps 4-6Mbps 1.5Mbps 176x144 (QCIF) 176x144 (QCIF) 352x288 (CIF) 352x288 (CIF) 176x144 (QCIF) 352x288 (CIF) 720x480 (ITU-R 601) 1920x1080 720x480 (ITU-R 601) 1920x1080 352x240 (SIF) Typical Associated Audio Quality Speech Speech Speech, Music Stereo CD, Surround Sound Surround Sound Stereo CD Quality
MPEG Applications MPEG-1 MPEG-2 MPEG-4 H.264 H.264 SE CD- ROM DVD DTV PDA Handphone PC DMB HD-DVD Scalability Homenetwork Streaming 1993 1995 1998 2004 2006 Scalability? 기존의비디오코딩 : 1 번의인코딩으로 1 가지의수신조건만만족 SVC: 1 번의인코딩으로여러해상도, 프레임율, 비트율복원가능 Multi-type network channel, multiple device 환경에최적
비디오데이터압축의핵심기술 기본원리 : 비디오데이터에존재하는공간적, 시간적, 그리고통계적중복성을인간시각적인특성을고려하여제거 관련기술 Prediction ( 예측 ) Transformation ( 변환 ) Quantization ( 양자화 ) Entropy Coding ( 엔트로피부호화 ) Hybrid Coding (MC + T + VLC)
영상압축표준의범위 source Pre-processing Encoding 저장 / 전송 destination Pre-processing Decoding
MPEG-1,2 Video Structure (1) 6
MPEG-1,2 Video Structure (2) 7
MPEG-1,2 ENCODER QP input image - + DCT Quantization VLC buffering 1 2 3 4 De-Quantization IDCT Motion Estimation and Compensation 5 + 8
움직임보상 (Motion Compensation) 각프레임의영상공간을 NxN (luminance 영상은 N=16, 4:2:0 의부표본화가적용되었을때 chrominance 영상은 N=8) 의매크로블록 (macroblock) 단위로분할 현재의프레임을타깃프레임 (target frame) 으로설정 타깃프레임의매크로블록과이전또는이후프레임 ( 참조프레임 (reference frames)) 들사이의정합을계산한다. 타깃매크로블록과참조프레임의가장잘정합된매크로블록의변위가움직임벡터 (motion vector) 이다. 전방향예측 (forward prediction): 참조프레임이이전영상인경우후방위예측 (backward prediction): 참조영상이미래영상인경우 움직임벡터탐색영역 : 타깃매크로블록을중심으로 (2p+1)x(2p+1)
Block Transform Encoding Original image 139 144 149 153 144 151 153 156 150 155 160 163 159 161 162 160 DCT 1260-1 -12-5 -23-17 -6-3 -11-9 -2 2-7 -2 0 1 quantise 79 0-2 -1-1 -1 0 0-1 0 0 0 0 0 0 0 run length code 0 79 1-2 0-1 0-1 0-1 2-1 0 0 Zig-zag Huffman code 79 0-1 0-2 -1 0 0-1 -1 0 0 0 0 0 0 10011011100011.
비디오압축의특성파라메터 Throughput of the channel: Data rate + ECC Distortion of the decoded video Delay (latency) Complexity Problem Formulation 최대허용 delay 와 complexity 하에서 bit-rate 와 distortion 의 Trade-off 최적화
MPEG-2 SYSTEM
MPEG-2 SYSTEM (TS) (1)
PES packet Header MPEG-2 SYSTEM (TS) (2) PES packet payload Header Payload Adaptation Field (used for stuffing here)
MPEG-2 SYSTEM (TS) (3)
Program Stream MPEG-2 SYSTEM (PS) Primarily intended for storage and retrieval from storage media Grouping of video, audio, and data elementary streams that have a common time base Each program stream consists of only one program Useful in error free environments Large packet size Packets size may be variable (hard for decoder to predict start and end of packets) DVD standard uses the MPEG-2 Program Stream PES (PSM) PES (PSD) PES ( 오디오 ) PES ( 비디오 ) PES 팩헤더팩페이로드팩헤더팩페이로드 PSM (Program Stream Map) : 프로그램을구성하는비디오와오디오의속성정보 PSD (Program Stream Directory): 임의엑세스가가능한엑세스단위의주소정보및표시시각을저장한표
핵심기술의압축성능비교 새로운기능을추가할때복잡도의증가나추가정보를나타내야할비트수증가에대해주의요망 IEEE Proceedings, pp.18-31, Jan 2005 Case 1) Spatial-transform + Intra coding only (e.g., JPEG) Case 2) Adding Skip mode to form a CR coder Case 3) Adding residual difference coding, but with only zero-valued MVs Case 4) Adding integer-precision MC with blocks of size 16x16 luma samples Case 5) Adding half-sample-precision MC Case 6) Allowing some 16x16 regions to be split into four blocks of 8x8 luma samples each for MC Case 7) Increasing MV precision to quarter-sample
MCP (Motion-Compensated Prediction) 항목내용비고 Fractionalsample-accuracy 영상경계에서의 MV Bipredictive MCP Variable block size MCP Multipicture MCP 정수픽셀단위이하의 MV 정밀도 픽셀간가상의픽셀은보간 공간정밀도가높을수록 MV 의정확도향상 영상경계확장 (extrapolation) 2 개의 MV 의평균치 Uncovered 영역이나평탄하고지속적인움직임의경우유리 MV 당픽셀의면적 움직임필드의정확도와 MV 를표현하기위한비트수사이의 trade-off MV 를검출하기위해참조하는앞뒤영상의수 많을수록장거리통계적상관성반영 ½ 샘플 : MPEG-1,2, H.263 ¼ 샘플 : MPEG-4 ASP (luma), H.264 1/8 샘플 : H.264 Chroma H.263 부터채택 MPEG-1 이후모든표준 H.264 에서 16x16 이하의움직임벡터단위 (macroblock) 허용 H.264 에서 5 개까지의복호화된참조프레임허용
H.264/AVC 개요 main goals - enhanced compression performance - network-friendly video applications Features - 4 x 4 integer DCT (avoid mismatch problem) - variable block-size motion compensation - quarter-sample-accuracy motion compensation - multiple reference frame motion compensation - directional spatial prediction for intra coding - more effective entropy coding context-adaptive variable-length coding (CAVLC) context-adaptive binary arithmetic coding (CABAC) - more flexible encoding features flexible slice size flexible macroblock ordering (FMO) arbitrary slice ordering (ASO)
MPEG-4 와 H.264/AVC Functionalities MPEG-4 Visual H.264 Supported data type Rectangular video frames and fields, arbitrary-shaped video objects, still texture and sprites, synthetic or synthetic-natural hybrid video objects, 2D and 3D mesh objects Rectangular video frames and fields Number of profiles 19 3 (+ High Profiles for FRExt) Compression Efficiency Medium High Support for video streaming Motion compensation minimum block size Scalable coding 8 x 8 4 x 4 MV accuracy ½ or ¼ ¼ Switching slices Transform 8 x 8 DCT 4 x 4 DCT approximation Built-in deblocking filter No Yes
H.264 구조 다양한전송및저장시스템을고려 Flexibility & Customizable
NAL Structure 항목 NAL Unit Paramet er Sets Access Units 기능 부호화가완료된비디오데이터는정수개의바이트단위인 NAL Unit 로패킷화됨 NAL 의첫번째바이트는데이터형을나타내는헤더 VCL NAL Unit 비디오데이터포함 Non-VCL NAL Unit 파라메타나 enhancement 정보를위한보조데이터 손실에대한강인성을위해중요한데이터 (MV 값및 type) 와덜중요한데이터 (residual DCT 계수 ) 를서로다른 NAL Unit 에배치하여서로다른중요도로전송 데이터 partitioning Sequential Parameter Sets (SPS): 부호화된일련의비디오픽쳐에적용 Picture Parameter Sets (PPS): 하나혹은그이상의개별픽쳐의복호에적용 SPS 와 PPS 의분리로 robustness 증대 Robustness 를위해 SPS 와 PPS 를반복적으로전송가능 단일복호된픽쳐와관련된 VCL 및 non-vcl 의셋을말함 픽쳐의모든 MB + 에러극복을위한추가데이터 (redundant slices)
H.264 구조 ( 비디오포맷 ) 비디오포맷 4:2:0 순차및비월주사비디오지원 비월주사관련모드 Frame mode: 두개의필드를하나의프레임으로통합하여하나의픽쳐로부호화 Field mode: 각필드를독립된픽쳐로간주 Macroblock-adaptive frame/field mode (MBAFF) 전체프레임을하나의픽쳐로부호화하지만 Prediction 과 Residual 부호화를위해수직방향으로인접한두개의 MB 를선택적으로두개의필드로분리가능 - 위의세가지모드의선택은각프레임별로적응적수행 - 첫번째두개의모드만을선택하는경우 Pictureadaptive frame/field (PAFF) 라부름 - 필드모드인경우 : MCP 는참조필드사용, 계수스캔이지그재그와다름, 수평에지에강한 de-blocking 사용안함
H.264 구조 (Slices) Slices 픽쳐내의 raster scan 방향의 MB 의집합임. 영상복호화의기본독립단위 error resilience (resynchronization) - Error resilience 기능은 FMO (Flexible Macroblock Ordering) 기술을사용하여더강화시킬수있음 - FMO 는픽쳐내의 MB 를복수의집단으로나누어각각을다른슬라이스에배정함 ( 예, foreground slice group and leftover, checkerboard 등 ) - ASO (Arbitrary Slice Ordering): FMO 와관련된것으로 ASO 가사용되지않으면손실강인성과지연감소를위해슬라이스순서대로첫번째 MB 은증가하는순서로배열되어야함 슬라이스들사이의독립성을이용하여병렬처리가능.
Slice Syntax H.264 구조 (Slices)
H.264/AVC CODEC Structure Basic structure of H.264/AVC for a macroblock
MC/ME Input Video Signal Split into Macroblocks 16x16 pixels - Decoder Coder Control Transform/ Scal./Quant. Scaling & Inv. Transform Control Data Quant. Transf. coeffs Entropy Coding Intra/Inter Intra-frame Prediction Motion- Compensation De-blocking Filter Output Video Signal Motion Estimation Motion Data
Inter Frame Prediction (P Slices) 다양한블록사이즈 ¼sample accuracy - 6 tap FIR filtering to ½ sample accuracy ( 수평및수직방향 ) - simplified filtering ( 정수위치와 ½ 위치의값을평균 ) to ¼ sample accuracy Picture boundary 에서도 MV 계산 MV 는이웃블록으로부터의예측이나중간값으로차분부호화 Multiple reference pictures P_Skip 모드 : 양자화된예측에러도 MV 도보내지않음, index 0 (list 0) 에위치한 MB Types 8x8 Types 16x16 0 8x8 0 16x8 0 1 8x4 0 1 8x16 0 1 4x8 0 1 Motion vector accuracy 1/4 8x8 0 1 2 3 4x4 0 1 영상을참조로 P_16x16 처럼복원 변화가없거나일정한움직임 (slow panning) 영역에서유리 2 3 8x8 이선택되면각 8x8 블록마다그것이더작은블록 8x4, 4x8, 혹은 4x4 로나누어지는지나타내는추가적인신택스필요
Motion Compensation Accuracy
Multiframe Motion Compensation
Multiframe Motion Compensation ( 계속 ) 전방향참조프레임최대 5 장후방향 1 장 Reference Picture ListX - Encoder 와 Decoder 에함께존재 - 전방향 List0, 후방향 List1 - DPB(Decoded Picture Buffer) 에 Reference Picture 저장 - Short term or long term reference picture Multiframe 의장단점 장점 - 반복적인움직임에효과적 - 객체의움직임으로가려진영역에효과적 - 부화소단위의움직임에효과적 - 주기적인빛의변화에효과적 - 바로이전의영상에잡음이있는경우효과적 단점 - 인코더복잡도증가 - 많은메모리요구
Intra Prediction Input Video Signal Split into Macroblocks 16x16 pixels - Decoder Coder Control Transform/ Scal./Quant. Scaling & Inv. Transform Control Data Quant. Transf. coeffs Entropy Coding Intra/Inter Intra-frame Prediction Motion- Compensation De-blocking Filter Output Video Signal Motion Estimation Motion Data
Intra Prediction ( 계속 ) Two modes for luma block - Intra 4x4 : 9 modes, texture area - Intra 16x16 : 4 modes, flat area -I_PCM One mode for chroma block - Similar to intra 16x16 Intra 16x16
Intra 4x4 Intra Prediction ( 계속 )
Intra Prediction ( 계속 ) 모든슬라이스 type 에대해 2 가지의기본적인 Intra Prediction: Intra_4x4, Intra_16x16 ( 특별한경우 I_PCM 이라불리는 3 번째의인트라코딩존재 ) Intra_4x4 는각 4x4 luma block 에사용되며상세한영상내용부위에적합 Intra_16x16 은상당히평탄한영상부위에적합 공간영역에서의 Intra coding 이이웃의블록에 Inter prediction 과함께쓰일때에러가시공간적으로전파될수있으므로 constrained Intra coding mode 를사용할수있음. Intra_16x16 mode: 16x16 luma 성분이일시에예측됨 Intra_16x16 mode 의 4 가지 mode: vertical, horizontal, DC, plane plane mode: position-specific linear combination, 천천히변하는영역에유리 Intra MB 의 chroma sample 은 luma 의 Intra_16x16 과유사 I_PCM Intra MB type 은 prediction 이수행되지않고보상없이원데이터가그대로전송됨 1) 부호화를위해필요한비트수가원데이터의비트수보다많을수없다, 2) 특정부위의 Lossless 부호화효과 Rate distortion optimization (RDO) 을위한모든가능한모드의수는 4 x (9 x 16 + 4) = 592 different RDO calculations
Rate-Distortion Optimization 최적 Intra & Inter Prediction Mode 와블록사이즈의결정을위해 Rate-Distortion Optimization (RDO) 알고리즘사용 Rate and distortion are function of QS. Rate and distortion are also functions of the percentage of null quantized coefficients ( zeros ) in the block.
Transform Coding Input Video Signal Split into Macroblocks 16x16 pixels - Decoder Coder Control Transform/ Scal./Quant. Scaling & Inv. Transform Control Data Quant. Transf. coeffs Entropy Coding Intra/Inter Intra-frame Prediction Motion- Compensation De-blocking Filter Output Video Signal Motion Estimation Motion Data
Transform Coding ( 계속 ) DCT 의역변환이정수계산이므로역변환 mismatch 문제해소및복호화복잡도를낮춤 DCT 블록의양자화계수는 zig-zag 스캔 Chroma 의 2x2 DC 계수는 raster-scan 4x4 Block Integer Transform Repeated transform of DC coeffs for 8x8 chroma and 16x16 Intra luma blocks (Intra_16x16) Hadamard Transform for luma Hadamard Transform for chroma
Quantization 모든 DCT 계수들은스칼라양자화됨 Logarithmic step size control Extended range of step sizes Smaller step size for chroma (cf. H.263 Annex T) Table-driven: 12.5% increase in qstep per 1-QP increase Reconstruction is 16-bit multiply, add, shift
DC 계수 (-1, 16, 17) 우선전송 Transmission Order
Entropy Coding Input Video Signal Split into Macroblocks 16x16 pixels - Decoder Coder Control Transform/ Scal./Quant. Inv. Scal. & Transform Control Data Quant. Transf. coeffs Entropy Coding Intra/Inter Intra-frame Prediction Motion- Compensation De-blocking Filter Output Video Signal Motion Estimation Motion Data
CAVLC Preamble: # of nonzero coefficients, # of T1 (Trailing 1 s) (4 개의 VLC 테이블중한개사용 ) Significant coefficient 들의부호와레벨값이스캔의역순으로부호화 ( 각레벨값을부호화할때 6 개의테이블중하나를선택하여이전에레벨에따라부호화됨 ) 마지막 non-zero 값이전의 zero 의수 각 significant level 당이전에발생한연속적인 zero 의수 각부호화단계마다최대가능 zero 의값에따라 run 값의부호화를위한테이블선택 전체 32 개의 VLC 테이블사용 테이블의일부는저장공간이필요없이 on-line 계산에의해결정 기존의 run-length 부호화에비해 2-7% 의비트율절감
CAVLC ( 계속 )
De-blocking Filter Input Video Signal Split into Macroblocks 16x16 pixels - Decoder Coder Control Transform/ Scal./Quant. Scaling & Inv. Transform Control Data Quant. Transf. coeffs Entropy Coding Intra/Inter Intra-frame Prediction Motion- Compensation De-blocking Filter Output Video Signal Motion Estimation Motion Data
De-blocking Filters ( 계속 ) De-blocking 필터를 Coding Loop내에놓은이유 : (1) MC의참조영상을모두필터링된깨끗한영상으로사용 (2) 복호기에 De-blocking 필터를의무적으로설치토록하여최소화질보장 3개의레벨에따른적응처리 (1) On Slice Level: global strength가비디오시퀀스의개별특성에의해조절 (2) On block edge level: inter/intra 예측결정, 움직임차이, 두개의참여블록의부호화된 residual의존재여부에따라필터강도조절 (0~4) (3) On sample level: 모든에지를가로지르는샘플값들을분석하여실제에지인지블록경계인지구분 De-blocking 필터는필터처리를하지않은것에비해객관적인화질을유지하면서 5-10% 의비트율을줄임. ( 주관적화질향상은더큼 )
De-blocking Filters ( 계속 ) Deblocking filter: Highly compressed decoded inter picture 1) Without Filter 2) with H264/AVC Deblocking
H.264 프로파일 High Compression quality, e.g. broadcast, but high complexity Extended Profile Streaming SP/SI Slices Data Partitioning MB-level frame/ Filed switching B Slices Weighted Prediction Interlace I/P Slices De-blocking filter CAVLC Multiple Ref. Frames ¼ sample MC Intra Predict. FMO and ASO Redundant Slices Baseline Profile CABAC Main Profile High Profile Adaptive Tr 4x4 or 8x8 HVS weighting matrices Predictive Lossless Monochrome format Low complexity and low delay: Lower capability plus error resilience (video conferencing wireless)
FRExt Amendment Fidelity Range Extensions (FRExt) Amendment 1: 2004 년 7 월완성 Content-contribution, content-distribution, Studio editing and postprocessing 의응용을위해 - Use more than 8 bits per sample of source video accuracy - Use 4:2:2 or 4:4:4 -Alpha-blending - Use very high bit-rates - Use very high resolution - Achieve very high fidelity even representing some parts of the video lossless - Avoid color-space transformation rounding error - Use RGB color representation Produce the High profiles: support all features of the prior Main profile and additionally support an adaptive transform coding and perceptual quantization scaling matrices Applications: HD-DVD, BD-ROM, DVB standards for European broadcasting television
H.264 Levels (1) 레벨 주어진프로파일내에서도모든가능한신택스를사용할수있도록구현하는것은비경제적이므로각프로파일당레벨 (Level) 을정의 레벨은비트스트림내의신택스요소의값에대한제한 ( 한정 ) H.264/AVC 에각프로파일에대해 15 개의레벨이정의됨 각레벨은비트스트림의상한치혹은디코더처리율의하한치를규정 -QCIF 로부터 4kx2k 의영상의크기 - 디코더처리율 : 초당 1485 에서 983040 블록처리 - 움직임벡터범위 : [-64,+63.75] 에서 [-512,+511.75]
H.264/AVC Levels (2)
Complexity of Codec Design Codec design includes relaxation of traditional bounds on complexity (memory & computation) rough guess 2-3x decoding power increase relative to MPEG-2, 3-4x encoding Problem areas: Smaller block sizes for motion compensation (cache access issues) Longer filters for motion compensation (more memory access) Multi-frame motion compensation (more memory for reference frame storage) More segmentations of macroblock to choose from (more searching in the encoder) More methods of predicting intra data (more searching) Arithmetic coding (adaptivity, computation on output bits)
Error Detection for H.264/AVC Constraints imposed on the H.264/AVC video bitstream syntax 1) Codeword for the VLC code, the transform coefficient, the motion vector code, CBP, DQUANT, the MBTYPE code, or the REFFRAME code. 2) The total number of decoded MBs in a slice should be equal to the size of the slice. 3) The number of the decoded transform coefficients within a 4x4 block should be smaller than 16. 4) Invalid video data are detected. For example, the prediction error between the predictive block and the current block is an invalid value.
Comparisons (Coding Efficiency)(1) Video Streaming Applications 352x288 progressive A. Joch et al., Performance comparison of video coding standards using Lagrangian coder control, ICIP 2002
Comparisons (Coding Efficiency)(2) Video Conferencing Applications 352x288 progressive A. Joch et al., Performance comparison of video coding standards using Lagrangian coder control, ICIP 2002
Comparisons (Coding Efficiency)(3) Entertainment Applications 720x576 Interlaced
Comparisons (Coding Efficiency)(4) Digital TV Applications 1280x720 progressive G.J.Sullivan, et al., SPIE August 2004
Performance of FRExt High Profile 1920x1080 Progressive 24frames/s Blue-ray Disk Association
Comparisons (Hardware Complexity)(1) 기능 Variable Block Size Hadamard Transform RD-Lagrangian OPtimization Encoder Complexity 복잡도 메모리엑세스빈도를선형적으로증가시킴 이기능을사용하면화질을유지하면서비트율을약 4-20% 감축 추가적인모드마다복잡도가 2.5% 수준으로선형증가하나해당압축이득은포화 엑세스주기를약 20% 증가 화질대비트율의변화는작음 데이터전송증가 120% PSNR 향상 0.35dB 까지 비트율 9% 절략 B-frames 엑세스주기 16 에서 +12% 비트율 10% 까지감축 CABAC 엑세스주기 25-30% 증가 ( 단일가역 VLC 데이블채택시에비해 ) 비트율 16% 까지감축 Displacement vector resolution Search Range Multiple Reference Frames Deblocking Filter ¼pel position 을안쓰고 ½ pel 만쓰는경우엑세스주기와처리시간약 10% 감축 ¼pel MB 을사용하는경우아주낮은비트율을제외하고약 30% 까지부호화효율증가 참조프레임수와탐색영역을모두증가시키면엑세스주기가약 60 배까지증가 반면 PSNR 과비트율성능에는영향이미약함 프레임의수가증가할수록약 25% 의복잡도선형증가 증간이하의비트율에서는무시할정도 ( 약 2% 이하 ) 의비트율이득 높은비트율에서는좀더많은비트율이득달성 부호화복잡도에대한영향은거의없음 주관적인화질을상당히향상시킴
Comparisons (Hardware Complexity)(2) 기능 Decoder Complexity 복잡도 CABAC 엑세스주기 12% 까지증가 ( 단일가역 VLC 에비해 ) 비트율이높을수록증가율이높아짐 RD-Lagrangian OPtimization B-frames Hadamard Transform Deblocking Filter Displacement vector resolution 중간및낮은비트율에서평균약 5% 복잡도증가 높은비트율에서는영향력미약 11-29% 복잡도증가 첫번째 B-frame 의경우낮은비트율에서 50% 의추가복잡도증가, 중간에서고비트율에서 20-35% 두번째 B-frame 이후의추가시간증가는몇 % 에불과 부호화시간은약 5% 증가 메모리엑세스는무시할정도 필수적채택사양 복호기엑세스주기 6% 까지증가 Encoder 가 ½ pel position 의 MV 만보내면부호화시간은약 15% 감소함 상대적으로 H.264/AVC Encoder 의복잡도는 MPEG-4 Part 2 (SP) 의약 10 배이상 Decoder 는약 2 배이상증가 H.264/AVC Profile 의 Tool 과파라메터값의선택에따라압축성능을유지하면서하드웨어복잡도를 Encoder 6.5 배 Decoder 1.5 배까지줄일수있음
인코더의입력파라메터와성능지수관계 파라메터사용용도주의 Hadamard on/off Rd-optimization on/off Reference Frame FMO/ASO Variable Block Size Motion vector resolution 움직임벡터탐색영역 왜곡측정시 Hadamard 변환의사용여부 VLC 부호화를수행하고실제발생될비트량에따른 rd-cost 계산 Reference frame 의개수는압축의효율을높여주나회로의복잡도가높아짐 채널상에에러가발생할경우에러감춤을위한방법 16x16, 16x8, 8x8, 8x4, 4x8, 4x4 단위예측보상 ¼ pel 움직임벡터사용여부 움직임벡터의크기결정 약 20% 의복잡도증가 아주큰복잡도를갖고있기때문에실제하드웨어에적용한예는없음. 적용시최대 9% 의비트율향상 한프레임증가에대해 25% 의복잡도증가발생비트율은높은비트율에서 14% 의향상을가져옴 비트율향상없음디코딩효율을위한파라메터 각모드마다 2.5% 의복잡도증가비트율향상 4-20% 향상 ½ pel 움직임벡터경우보다 10% 복잡도증가저전송율시 30% 정도의비트율상승
H.264 Competitors Name Group Standard Strong Points Notes H.264 MPEG+ITU HD-DVD WMV9 (VC1) Microsoft DMB (main) HD-DVD (main) High performance (on low-bit rate) MPEG standard High performance (on low-bit rate) Moderate complexity Compatible to PC High Complexity MPEG-LA licence MS solution Patent issue CE vs IT issue AVS China EVD2 High performance (similar to H.264) moderate complexity Very large market China s Standard Patent issue WMV9 & AVS: Low complexity + Comparable performance to H.264
맷음말 H.264/AVC 의부호화툴은기존의 MPEG-4 와 MPEG-2 와비교해서약 50% 의비트절략이가능 그러나 H.264/AVC 의 Decoder 복잡도는 MPEG-4 Visual SP 에비해 2 배 Encoder 는약 10 배더복잡 H.264/AVC main profile decoder 는 MPEG-2 에비해약 4 배더복잡, Encoder 의복잡도는 rate-constrained encoder control 과움직임검출알고리즘에따라상당히다를수있음 VLSI 와 CPU 의성능향상과메모리의가격하락이 H.264/AVC 의구현을가능하게함 Reference S/W http://bs.hhi.de/~suehring/tml/download