臺灣博碩士論文加值系統

H.26L 是新一代的視訊編碼標準，它提供極低位元率、即時、低點對點延遲等應用。H.26L 意欲取代現存的視訊標準 H.263 與 MPEG4 ，以提供良好的壓縮效率 (比 H.263 節省 50% 的位元率)與進階功能(例如抵抗錯誤能力).從我們的複雜度分析報告得知，H.26L 視訊壓縮標準在於動態搜尋中使用7種不同的區塊其得到良好的壓縮率能，但是相對的，它也是最運算複雜度最高的地方。為了加快區塊比對速度，我們在此使用有效率的方法以減少其運算量。
對於動態搜尋演算法中，為了保持H.26L 良好的壓縮視訊品質，我們採用快速完全搜尋演算法，分別是SEA、PNSA與PPNM 等演算法，並且利用H.26L 在動態搜尋中其相同區塊與不同區塊之間七種不同搜尋區塊的高重疊特性，在每編每一張畫面之前，事先將 PPNM 中 所使用的1x4 區塊長度對於畫面中的每一點加以儲存，以減少運算量。根據我們的實驗結果，PPNM + PNS 的技巧在搜尋範圍 之中，可以比完全搜尋法大約快 3.3 倍左右，並且其搜尋速度隨著搜尋範圍增加而上升。
最後我們將 H.26L 編碼器移植到 TI TMS320C6701 DSP 平台中，並分別對其進行記憶體與程式碼最佳化。在程式碼最佳化中，對於整數轉換與Hadamard轉換在平行組語之中使用一些技巧，使得其速度能比C語言編譯器之結果分別快二倍與四倍多。對於 Sub-QCIF 格式之影像，TI TMS320C6701 每秒大約可編 0.21 張左右。

H.26L is a new, emerging video coding standard. It is aimed at very low bit rate, real-time, low end-to-end, and mobile application. H.26L intended to replace the previous H.263 and MPEG4 standards by targeting far beyond in compression performance and advanced features like error resilience. From the report [20] we know that the H.26L standard derives most of its performance gains from improved motion estimation. On the other hand, motion estimation is the most computation-intensive routine.
We adopt fast-full search algorithm SEA, PNSA and PPNM and make use of overlaps of search range of the seven different block size for motion estimation. Before encoding one frame, we calculate and save 1x4 partial norm in advance to consist of seven different block sizes used in PPNM algorithm. Operation can be reduced while reusing 1x4 partial norm in PPNM algorithm, especially when the search range increases. According to our experimental results, the PPNM with PNS algorithm is faster about 3.3 times than full search within search range.
Finally, we port H.26L encoder from PC to TI TMS3206701 DSP platform. And we use techniques in parallel assembly code for integer transform and Hadamard transform to enhance encoding process. For Sub-QCIF format image, TI TMS320C6701 running on 100MHz can encode about 0.21 frame per second.

中文摘要 I
Abstract ii
誌謝 iii
Contents iv
List of Figures vi
List of Tables vii
Chatper 1 An Introduction to H.26L 1
1.1 Overview 1
1.2 Intra Prediction and Coding 2
1.3 Inter Prediction and Coding 3
1.3.1 Block sizes 3
1.3.2 Motion Estimation Accuracy 4
1.3.3 Multiple reference picture selection 4
1.3.4 De-blocking filter 4
1.4 Integer Transform 5
1.5 Quantization and Transform Coefficient Scanning 5
1.6 Entropy Coding 6
1.6.1 Universal VLC 6
1.7 Performance Analysis 6
Chatper 2 Optimization for Motion Estimation In H.26L Encoder 7
2.1 Profile of H.26L Encoder with TML-5.9 7
2.2 Fast Full-search Algorithm 10
2.2.1 Successive Elimination Algorithm 11
2.2.2 Progressive Norm Search Algorithm 16
2.2.3 Progressive Partial Norm Matching 21
2.2.4 Progressive Partial Norm Matching with Partial Norm Saving 22
Chatper 3 Efficient Implementation of H.26L Encoder on TMS3206701 31
3.1 Introductions 31
3.1.1 The TMS320C6701 EVM 32
3.1.2 The C6x Compiler and Assembly Optimizer 32
3.2 H.26L Encoder 31
3.2.1 Analysis of H.26L Video Encoder on ‘C67x Platform 34
3.2.2 Code Optimization 35
3.2.2.1 Optimizing SAD Computations for Motion Estimation 36
3.2.2.2 Integer/inverse Transform 37
3.2.2.2 Hadamard Transform 38
3.2.2 Memory Optimization 41
3.3 Overall Performance of H.26L Video Encoder on the ‘C67x 43
Chatper 4 Conclusion and Future Works 46
4.1 Conclusions 47
4.2 Future Works 48
Bibliography 49
Appendix A 5

[1]LI. W. and SALARI. E :“Successive elimination algorithm for motion estimation”
, IEEE Trans, Image Process, 1995, 4, pp.105-107
[2]OH. T.M., KIM. Y.R, HONG. W.G, and KO. S. J: “Partial norm based search algorithm for fast motion estimation”, Electron Lett, 2000,36,(14),pp.1195-1196
[3]W. -G. Hong, T. -M. Oh and S. -J.Ko : “Fast motion-estimation algorithm based on
progressive partial norm matching”, Electron Lett, 2001,36,(14),pp.890-892
[4]TI Technique Document: “SPRA189 TMS320C6000 CPU and Instruction Set Reference Guide”, 1999
[5]TI Technique Document: “SPRU187 TMS320C6000 Optimizing Compiler”, 1999
[6]TI Technique Document: “SPRU198 TMS320C6000 Programmer’s Guide”, 1999
[7]A. Murat Tekalp, university of Rochester. :“Digital Video Processing”, Prentice Hall PTR Upper Saddle River, NJ 07458, 1995.
[8]Shan Zhu and Kai-Kuang Ma. “A New Diamond Search Algorithm for Fast Block-Matching Motion Estimation,” IEEE Transactions on Image Processing, Vol.9 Issue: 2, pp.287 —290, Feb 2000.
[9]Yao. W., Jorn O., Ya-Qin. Z.: “Video Processing And Communications”, Prentice Hall PTR Upper Saddle River, 2002.
[10]Yuji I.: ”Bi-directional motion vector coding using univeresal VLC”, Signal Processing: Image Communication, Vol, 14,pp.541-557, May 1999
[11]ITU-T Recommendation H.263, “Video coding for low bit rate communication”,
[12]ITU-T Q.16/SG16, “H.26L Test Model Long Term Number 5(TML-5) draft0”,
Sept 2000.
[13]Telenor Broadband Services, “TML H.26L Codec v.5.9”, via ftp://standard.pictel.
com/video-site/, Nov 2000.
[14]Bojkovic, Z.S, “MPEG and ITU-T video communication: standardization process”, Telecommunications in Modern Satellite, Cable and Broadcasting Services, 1999. 4th International Conference on , Volume: 1 , 1999
Page(s): 281 -290 vol.1
[15] Itoh, Y.; Ngai-Man Cheung, “Universal variable length code for DCT coding” , Image Processing, 2000. Proceedings. 2000 International Conference on , Volume: 1 , 2000 Page(s): 940 -943 vol.1
[16] TI Technique Document: “SPRU197 TMS320C6000 Technical Brief”, 1999
[17] TI Technique Document: “SPRU269 TMS320C6201/6701 Evaluation Module
User’s Guide”, 1999
[18] Ying-Jui Chen; Soontorn Oraintara; Truong Nguyen “Video compression using integer DCT”, Image Processing, 2000. Proceedings. 2000 International Conference on , Volume: 2 , 2000 Page(s): 844 -847 vol.2
[19] Lehtoranta, O.; Hamalainen, T.; Saarinen, J.,” Real-time H.263 encoding of QCIF-images on TMS320C6201 fixed point DSP”, Circuits and Systems, 2000. Proceedings. ISCAS 2000 Geneva. The 2000 IEEE International Symposium on , Volume: 1 , 2000 Page(s): 583 -586 vol.1
[20]Lappalainen, V.; Hallapuro, A.; Hamalainen, T.D. “Optimization of emerging H.26L video encoder “, Signal Processing Systems, 2001 IEEE Workshop on, 2001 Page(s): 406 —415