http://dx.doi.org/10.5909/jeb.2012.17.1.165 a), a), a) Motion Vector Coding Using Adaptive Motion Resolution Myung-Hun Jang a), Chan-Won Seo a), and Jong-Ki Han a) MPEG-2, MPEG-4. KTA...,. Threshold. HEVC HM3.0, Random Access 0.9%, Low Delay B picture 0.6%, P picture 2.7%. Abstract In most conventional video codecs, such as MPEG-2 and MPEG-4, inter coding is performed with the fixed motion vector resolution. When KTA software was developed, resolution for MVs can be selected in each slice. Although KTA codec uses a variety of resolutions for ME, the selected resolution is applied over the entire pixels in the slice and the statistical property of the local area is not considered. In this paper, we propose an adaptive decision scheme for motion vector resolution which depends on region, where MV search area is divided to multiple regions according to the distance from PMV. In each region, the assigned resolution is used to estimate MV. Each region supports different resolution for ME from other regions. The efficiency of the proposed scheme is affected from threshold values to divide the search area and the entropy coding method to encode the estimated MV. Simulation results with HM3.0 which is the reference software of HEVC show that the proposed scheme provides bit rate gains of 0.9%, 0.6%, and 2.9% in Random Access, Low Delay with B picture, and Low Delay with P picture structures, respectively. Keyword: Motion Vector Resolution, Motion Estimation, Motion Vector Coding a) Sejong University, Dept. of Information and Comm. Engineering : (hjk@sejong.ac.kr) (). [KI002142, ] (20111214),(20111221),(20111221)..
,.. H.261, MPEG-2 MPEG-4 [1-3]. H.264/AVC - 1/4 [4-6]. H.264/AVC, H.264/AVC ITU-T VCEG Key Technical Area (KTA) [7]., 1/8 1/4 [8]. High Efficiency Video Coding (HEVC) H.264/AVC 1/4 [9]..,,. 1/8,..,..,.,, Threshold. Threshold. Threshold -.. II. III, Threshold IV. V, VI...,.,. H.264/AVC KTA HEVC HM3.0 [5][7][10]. (1)
.,.. Sum of Absolute Difference (SAD).,,. (1). (3).,..... 2, 3 Exponential-Golomb code., x 3, 5 00110 5. 1/8, 47 3 11.. 2. Table 2. Code number assignment according to MV resolutions 1 Exponential-Golomb code [6][7][9]. 1/8 1/4 1/2 1/1 1. Exponential-Golomb code Table 1. Exponential-Golomb code for fixed MV resolution 0 0 0 1 1 010-1 2 011 2 3 00100-2 4 00101 3 5 00110......... 0 0 0 0 0 1/8 1-1/8 2 2/8 3 1-2/8 4 2............... 8/8 15 7 2 1............... 24/8 47 23 6 3...............
3. Exponential-Golomb code Table 3. Exponential-Golomb code for code number 0 0 1 010 2 011 3 00100 4 00101 5 00110 6 00111 23 000011000 47 00000101111.. [11]. HEVC HM3.0 [10]. HM3.0 1/4. HEVC (412 240), HEVC [12]., 1/2 HM3.0, 4.. 1.,.,.,.. (1) 4. 1/2 Table 4. Performance with 1/2-pel MV resolution BD-Rate (%) (416 240) BasketballPass -0.1 BQSquare 9.4 Blowing Bubbles 2.7 RaceHorses 0.9 Average 3.2 4 1/4 1/2 3.2%., BQSqure Blowing Bubbles., BasketballPass 1/2 0.1%..,
5 HM3.0 1/8. 5. 1/8 Table 5. Performance with 1/8-pel MV resolution BD-Rate (%) BasketballPass 1.9 BQSquare -5.7 Blowing Bubbles -0.4 1. 1/8.. RaceHorses 2.4 Average -0.5 5, HM3.0 1/4 1/8,,. 4 5,,..,,.,,.. 2. 1. Fig. 1. Ranges for MV resolutions based on absolute value of MVD 1 1/8 1/4, 1/2,, Threshold. 1,. HM3.0 2.
2. Fig. 2. Conventional code number assignment for MVDs 3. Fig. 3. Proposed code number assignment for MVDs HM3.0, 3. 3,, S 0, 1/8, 1/4, 1/2,..,. 24, 1, 47 3 11.,, 17. x y. x y.. 4 4. Fig. 4. MV resolution decision based on absolute value of MVD
, x y, x. 4 x 1/2.,. 3.,.,.. i f i f i f,. 6. 6. Table 6. MV resolution and index (r) 1/8-pel 0. Threshold Threshold. 3. 1 2 Threshold. Threshold. 1. 1/8.,., Prediction Unit (PU).. 1 PU 2 Threshold. Threshold {,,, } {0, max, max, max }. max. Threshold {,,, } {0, 2, 4, 8} 1/8, Threshold. Threshold, 0, 1., Threshold. 1/4-pel 1 1/2-pel 2 1-pel 3 (6) Threshold
1 PU., Threshold. PU, (8). 1, PU. (8) Threshold. (6)(8) 1.,,. Threshold,. arg min Threshold. (6)(11) Threshold. 5 Threshold. Th = { Th0, Th1, Th2, Th3} = { 0, SRmax, SRmax, SRmax } Th { Th0, Th1, Th2, Th3} = { 0,2,4,8} k = 0, r = 1 D = D D D D Th = D Th k r r arg min (9) Threshold. (10), 3 Threshold,, 1, Threshold. Threshold { } * m = arg min m D Thr m ³ Thr - 1 k = m * T S J k J j j= 1 ( ) = å Th ( ) r ( 1) max D Th k + > SR ( ) * k = arg min J S k Th = D Th k r r r ³ 3 5. Threshold Fig. 5. Block diagram for optimal threshold decision *
3 4 Threshold., Threshold. Threshold, (12). (12).. 8. DCT-IF 8 taps () Table 8. Coefficients of DCT-IF 8 taps (Chroma component) Position Filter Coefficients 1/16 {-2 63 4-1} 3/16 {-5 59 13-3} 5/16 {-6 52 23-5} 7/16 {-7 43 34-6} 9/16 {-6 34 43-7} 11/16 {-5 23 52-6} 13/16 {-3 13 59-5} 15/16 {-1 4 63-2}.. HEVC HM3.0, JCT-VC [10][12]. HM3.0 1/8, 7 8 DCT-IF 8tap [13-14]. BD-Rate,,,. 916 RA Random Access, LB Lowdelay with B picture, LP Low dealy with P picture, HE High Efficiency LC Low Complexity 7. DCT-IF 8 taps () Table 7. Coefficients of DCT-IF 8 taps (Luma component) Position Filter Coefficients 1/8 {-1, 3, -6, 62, 9, -4, 2, -1} 3/8 {-2, 5, -12, 50, 30, -10, 4, -1} 5/8 {-1, 4, -10, 30, 50, -12, 5, -2} 7/8 {-1, 2, -4, 9, 62, -6, 3, -1} 911 HM3.0. Random 9. Random Access Table 9. Performance for Random Access coding structure Class A (2560x1600) Class B (1920x1080) Class C (832x480) (416x240) RA-HE RA-LC BD-Rate (%) BD-Rate (%) Traffic -0.3-0.4 PeopleOnStreet -0.3-0.5 Nebuta 0-0.1 SteamLocomotive -0.9-1.0 Class A Average -0.4-0.5 Kimono -0.8-1.1 ParkScene -0.2-0.3 Cactus -0.4-0.5 BasketballDrive -0.3-0.5 BQTerrace -1.2-1.5 Class B Average -0.6-0.8 BasketballDrill -0.7-1.0 BQMall -0.5-0.6 PartyScene -1.2-1.9 RaceHorses -0.3-0.4 Class C Average -0.7-1.0 BasketballPass -0.3-0.5 BQSquare -3.1-5.0 Blowing Bubbles -0.8-1.3 RaceHorses -0.3-0.4 Average -1.1-1.8 Total Average -0.7-1.0
10. Low delay with B picture Table 10. Performance for Low delay with B picture coding structure Class B (1920x1080) Class C (832x480) (416x240) Class E (1280x720) LB-HE LB-LC BD-Rate (%) BD-Rate (%) Kimono -0.4-1.2 ParkScene -0.1-0.2 Cactus -0.2-0.2 BasketballDrive -0.3-0.6 BQTerrace -0.1-0.6 Class B Average -0.2-0.5 BasketballDrill -0.5-0.8 BQMall -0.3-0.4 PartyScene -1.1-1.9 RaceHorses -0.2-0.6 Class C Average -0.5-0.9 BasketballPass -0.2-0.5 BQSquare -3.3-5.7 Blowing Bubbles -0.6-1.0 RaceHorses -0.1-0.3 Average -1.1-1.9 Vidyo 1 0.5 0.2 Vidyo 2 0 0.6 Vidyo 3 0.2 0.4 Class E Average 0.2 0.4 Total Average -0.4-0.8 11. Low delay with P picture Table 11. Performance for Low delay with P picture coding structure Class B (1920x1080) Class C (832x480) (416x240) Class E (1280x720) LP-HE LP-LC BD-Rate (%) BD-Rate (%) Kimono -0.5-0.8 ParkScene -0.3-0.4 Cactus -0.5-0.4 BasketballDrive -0.4-0.4 BQTerrace -6.5-6.0 Class B Average -1.6-1.6 BasketballDrill -1.6-1.3 BQMall -2.2-2.9 PartyScene -7.0-6.5 RaceHorses -0.2-0.1 Class C Average -2.7-2.7 BasketballPass -0.3-0.3 BQSquare -16.7-16.1 Blowing Bubbles -3.7-3.6 RaceHorses -0.1-0.3 Average -5.2-5.1 Vidyo 1-0.8-0.2 Vidyo 2-3.9-0.7 Vidyo 3-1.5-0.3 Class E Average -2.1-0.4 Total Average -2.9-2.5 Access 0.9%, Low delay with B picture 0.6%, Low delay with P picture 2.7%. Low delay with B picture HM3.0 Generalized P and B pictures (GPB). GPB. GPB Low delay with P Picture GPB.., HM3.0 1/8., 1/8 7 8. 1214. HM3.0+1/8-pel HM3.0 1/8. HM3.0+1/8-pel Random Access 1.5%, Low delay with B picture 1.7%, Low delay with P picture 1.0%.
911. 1/8. HM3.0. 914 Random Access Low delay with B picture HM3.0 HM3.0+1/8-pel Low delay with P picture HM3.0+ 1/8-pel HM3.0. GPB. Random Access Low delay with B picture 1/8 1/4. Low delay with P picture 1/8 1/4. Low delay with P picture 1.0%. Threshold. 12. Random Access (vs. HM3.0+1/8-pel) Table 12. Performance for Random Access coding structure (vs. HM3.0+1/8-pel) Class A (2560x1600) Class B (1920x1080) Class C (832x480) (416x240) RA-HE RA-LC BD-Rate (%) BD-Rate (%) Traffic -0.9-1.0 PeopleOnStreet -2.7-3.3 Nebuta -0.5-1.0 SteamLocomotive -2.2-3.2 Class A Average -1.5-2.1 Kimono -2.2-2.7 ParkScene -0.9-1.1 Cactus -0.8-1.1 BasketballDrive -1.5-2.1 BQTerrace -0.8-1.1 Class B Average -1.2-1.6 BasketballDrill -1.2-1.7 BQMall -0.9-1.0 PartyScene -0.4-0.3 RaceHorses -2.0-2.7 Class C Average -1.1-1.4 BasketballPass -1.9-2.4 BQSquare -0.1-0.1 Blowing Bubbles -0.8-0.6 RaceHorses -2.5-3.1 Average -1.3-1.6 Total Average -1.3-1.7 13. Low delay with B picture (vs. HM3.0+1/8-pel) Table 13. Performance for Low delay with B picture coding structure (vs. HM3.0+1/8-pel) Class B (1920x1080) Class C (832x480) (416x240) Class E (1280x720) LB-HE LB-LC BD-Rate (%) BD-Rate (%) Kimono -1.6-2.9 ParkScene -1.0-1.8 Cactus -1.0-2.1 BasketballDrive -1.4-2.1 BQTerrace -0.7-1.6 Class B Average -1.1-2.1 BasketballDrill -1.9-1.9 BQMall -0.9-1.2 PartyScene -0.2-0.2 RaceHorses -1.8-2.8 Class C Average -1.2-1.5 BasketballPass -1.8-2.4 BQSquare 0 0 Blowing Bubbles -0.4-0.3 RaceHorses -2.3-2.8 Average -1.1-1.4 Vidyo 1-2.2-3.3 Vidyo 2-1.5-4.5 Vidyo 3-2.0-3.4 Class E Average -1.9-3.7 Total Average -1.3-2.1
14. Low delay with P picture (vs. HM3.0+1/8-pel) Table 14. Performance for Low delay with P picture coding structure (vs. HM3.0+1/8-pel) Class B (1920x1080) Class C (832x480) (416x240) Class E (1280x720) LP-HE LP-LC BD-Rate (%) BD-Rate (%) Kimono -1.5-2.4 ParkScene -0.3-0.9 Cactus -0.4-1.8 BasketballDrive -0.1-0.6 BQTerrace -0.1-0.2 Class B Average -0.5-1.2 BasketballDrill -0.9-1.4 BQMall 0-1.4 PartyScene 0 0 RaceHorses -0.8-1.4 Class C Average -0.4-1.0 BasketballPass -1.4-1.8 BQSquare 0.1 0.1 Blowing Bubbles 0-0.1 RaceHorses -1.9-2.3 Average -0.8-1.0 Vidyo 1-1.3-2.4 Vidyo 2-0.9-2.0 Vidyo 3-1.5-2.0 Class E Average -1.2-2.1 Total Average -0.7-1.3 15 16 HM3.0. Threshold HM3.0. 15 HM3.0 4. HM3.0 1/4 1/8 Threshold.. 16 HM3.0 2%., Threshold,.. HM3.0 2%. 15. Table 15. Encoder complexity of the proposed algorithm RA-HE (%) RA-LC (%) LB-HE (%) LB-LC (%) LP-HE (%) LP-LC (%) Class A 377 362 x x x x Class B 373 416 438 382 348 352 Class C 364 425 450 351 327 340 374 431 460 363 355 417 Class E x x 403 317 369 389 Average 373 408 440 357 349 371 16. Table 16. Decoder complexity of the proposed algorithm RA-HE (%) RA-LC (%) LB-HE (%) LB-LC (%) LP-HE (%) LP-LC (%) Class A 103 101 x x x x Class B 101 104 98 100 100 102 Class C 102 103 100 102 100 101 102 105 102 103 101 102 Class E x x 101 101 102 103 Average 102 103 99 101 101 102
..,. HM3.0 Random Access 0.9%, Low Delay B picture 0.6%, P picture 2.7%., DVD VOD. Threshold. [1] Committee Consultative International Telegraphie et Telegraph (CCITT), H261 video for audiovisual services at p*64 kbits, Geneva, 1990 [2] Int. Telecommun. Union-Telecommun. (ITU-T) and Int. Standards Org./Int. Electrotech. Comm. (ISO/IEC) JTC 1, Rec. H262 and ISO/IEC 13 818-2 (MPEG-2 Video), Generic Coding of Moving Pictures and Associated Audio Information-Part 2: Video 1994,11 [3] Int. Standards Org./Int. Electrotech. Comm. (ISO/IEC) JTC 1, ISO/IEC 14496-2 (MPEG-4 visual version 1), Coding of Audio-Visual Objects-Part 2: Visual 1999. [4] ITU-T Recommendation H.264 and ISO/IEC 14496-10 AVC Advanced video coding for generic audiovisual services, version 3: 2005 [5] T. Weigand, G. J. Sullivan, G. Bjontegaard, and A. Luthra, Overview of the H.264/AVC video coding standard IEEE Trans. Circuits Syst. Video Technol., Vol. 13, no. 7, pp. 560~576, 2003. 7. [6] Draft ITU-T Recommendation and Final Draft International Standard of Joint Specification (ITU-T Rec. H.264/IEC 14496-10 AVC). Mar, 2003 [7] ITU-T VCEG, Reference ITU-T VCEG-KTA Software is available at http://iphome.hhi.de/suehring/tml/download/kta [8] J.Ostremann and M.Narroschke, Motion compensated prediction with 1/8-pel displacement vector resolution, ITU-T SG16 Q6 VCEG-AD09, 2006, 10 [9] T. Wiegand, B. Bross, W.-J. Han, J.-R. Ohm, and G. J. Sullivan, WD3: Working Draft 3 of High Efficiency Video Coding, JCTVC-E603, Geneva, Switzerland, Mar, 2011 [10] K. McCann, B. Bross, S. Sekiguchi, W-J. Han, HM3: High Efficiency Video Coding (HEVC) Test Model 3 Encoder Description, JCTVC-E602. Geneva, Switzerland, Mar,. 2011. [11] M. H. Jang, C.W. Seo, and J. K. Han Motion Estimation using Region-based Adaptive Motion Vector Prediction, The 26th International Technical Conference on Circuits/Systems, Computers and Communications, Jun. 20-22 2011. [12] F. Bossen, Common test conditions and software reference configuration, JCTVC-E700, Geneva, Switzerland, Mar, 2011 [13] K. McCann, W-J. Han and I-K Kim, Samsung s Response to the Call for Proposals on Video Compression Technology, JCTVC-A124, Dresden, Germany, April 2010. [14] E. Alshina, J. Chen, A. Alshin, N. Shlyakhov, and W.-J. Han, CE3: Experimental results of DCTIF by Samsung, JCT-VC meeting contribution JCTVC-D344, Daegu, KR, Jan,. 2011.
178 방송공학회논문지 년제 권제호 2012 17 1 저자소개 장명훈 년 : 세종대학교 정보통신공학과 학사 년 ~ 현재: 세종대학교 정보통신공학과 석사과정 주관심분야 : HEVC, H.264/AVC, 영상처리 - 2010-2010 - 서찬원 년 : 세종대학교 정보통신공학과 학사 년 : 세종대학교 정보통신공학과 석사 년 ~ 현재 : 세종대학교 정보통신공학과 박사과정 주관심분야 : HEVC, H.264/AVC, SVC - 2007-2009 - 2010 - 한종기 - 년 : 한국과학기술원(KAIST) 전기 및 전자공학과 학사 년 : 한국과학기술원(KAIST) 전기 및 전자공학과 석사 년 : 한국과학기술원(KAIST) 전기 및 전자공학과 박사 년 2월 ~ 2001년 8월 : 삼성전자 DM 연구소 책임 연구원 년 9월 ~ 2006년 2월 : 세종대학교 정보통신공학과 조교수 년 3월 ~ 2011년 8월 : 세종대학교 정보통신공학과 부교수 년 9월 ~ 현재 : 세종대학교 정보통신공학과 교수 주관심분야 : HEVC, H.264/AVC, SVC, DVC, Transcoding, 영상처리, 신호처리 1992 1994 1999 1999 2001 2006 2011