臺灣博碩士論文加值系統

HEVC(High Efficiency Video Coding: 高效能視訊編碼)是近幾年興起的視訊編碼標準，與其前身H.264相比，HEVC可以得到更好的壓縮倍率，然而其代價就是必須花費大量的計算時間，我們利用市面上已在應用階段的x265測試，以解析度為1920x1080、畫面總數500張的作為測試影片，在QP=22的情況下仍然需要20分鐘的編碼時間。為了提升HEVC的編碼速度，提前終止演算法是目前大多數人研究的主題之一，其中大部分的提前終止演算法進行判斷的標準是閥值，但並非所有的影片都適合相同的閥值，所以若無法做到根據視訊自動調整閥值，都難以在HEVC上應用。為解決上述的情況，在本論文中，我們利用機器學習的方式來取代閥值，透過學習大量不同視訊的特徵，進而使電腦能夠在不同的視訊下，仍能成功判斷是否提前終止。
然而要使電腦能準確的判斷，除了大量的資料外，準確的特徵也是同樣重要的，如同人類判斷該物體是不是一輛車，不是從該物體的顏色做判斷，若我們將錯誤或不重要的特徵資訊交給電腦訓練，反而會使判斷失準，因此選擇訓練特徵是十分重要的。基於畫面內預測計算的主要參數是本身區塊的平滑度與相鄰區塊邊界的相似度，我們將這些資訊數值化後作為特徵，輸入至倒傳遞類神經網路(Backpropagation Neural Network; BPNN)模型中進行訓練，訓練完成後的類神經網路模型便可直接放入HEVC中進行提前終止的判斷。研究結果顯示，利用本篇論文所提出的提前終止演算法，可在BD-BR僅增加1.921%、BD-PSNR僅減少0.086dB的情況下，節省大約40. 784%的編碼時間。

HEVC (High Efficiency Video Coding) is the video compression standard that was developed in recent years. Compared with the former compression standard H.264, HEVC can make better compression ratio. But, the cost is that it takes a lot of computing time. An HEVC encoder is implemented on x265 to test video with resolution of 1920x1080 and 500 frames as a test sequence, in QP = 22 all intra mode, and it still needs almost 20 minutes to encode. To accelerate the coding of HEVC, early termination is one of the subjects studied by most people. Most of them decided whether early terminating the partition process or not by using threshold, but not all videos are suitable for the same threshold. Therefore, if it is impossible to automatically adjust the threshold in accordance with the input video, it is difficult to apply to HEVC with acceptable performance degradation. To solve this problem, machine learning approaches are employed to avoid the need of threshold, and by learning a huge amount of features of different videos, to help encoder successfully early decide coding unit partition in different videos.
However, in order to make accurate prediction, prominent features are equally important. Similar to people recognizing car objects, the color of the car will not be the key features for recognition. If the encoder trains the wrong or unimportant feature information, the decision will be incorrect. In intra prediction mode the block smoothness and similarity to the reference samples are used in the study. This feature set are transformed into normalized values, then feed to a Backpropagation Neural Network (BPNN) model to train a decision model. After training, the neural network model can be applied on HEVC encoder. Experiment result shows that the result in the proposed method can reduce about 40.784% of encoding time with 1.921% BD-BR increase and 0.086% BD-PSNR loss.

Abstract..........i
摘要..........iii
Acknowledgements..........v
Table of Contents..........vi
List of Tables..........viii
List of Figures..........ix
Chapter 1 Introduction..........1
1.1 Background..........1
1.2 Motivation and Objective..........2
1.3 Requirements..........3
1.4 Organization of the thesis..........3
Chapter 2..........High Efficiency Video Coding 5
2.1 HEVC Coding Structure..........5
2.2 HEVC Block Unit Type..........6
2.2.1 Coding Tree Unit..........6
2.2.2 Coding Unit..........7
2.2.3 Prediction Unit..........8
2.2.4 Transform Unit..........9
2.3 Rate-Distortion Cost..........9
2.3.1 Quantization..........10
2.3.2 Entropy Encoder..........10
2.4 Block Prediction Type─Intra Prediction..........10
2.4.1 Intra Prediction..........11
2.4.2 Reference Sample Generation..........12
2.4.3 Planner mode and DC mode..........14
2.4.4 Angular Mode..........15
2.4.5 RD cost and Intra Prediction..........15
2.5 x265..........17
2.5.1 Intra Prediction Process in x265..........18
2.6 Related Literature..........20
2.6.1 Fast Intra Prediction Mode Decision..........21
2.6.2 Fast Coding Unit Depth/Size Decision..........22
2.6.3 Summary..........25
Chapter 3 Backpropagation Neural Network..........26
3.1 Overview of Neural Network..........26
3.1.1 Neural Network Model..........27
3.2 Overview of Backpropagation Neural Network..........29
3.2.1 Multi-Layer Perceptron..........29
3.2.2 Backpropagation Algorithm..........30
3.2.3 Training..........32
3.2.4 Validation..........34
Chapter 4 BPNN with Texture Features based Early Termination Algorithm ..........35
4.1 The Concept..........40
4.2 Overview of Proposed Algorithm..........41
4.2.1 Pre-processing of Training Data..........41
4.2.2 Architecture of Proposed Method in x265 Intra Encoding..........42
4.3 BPNN Method Analysis..........43
4.3.1 BPNN with Block Texture Feature Analysis..........45
4.3.2 BPNN with Reference Sample Difference Feature Analysis..........46
4.3.3 BPNN with All Feature Analysis..........47
4.3.4 BPNN Method Summary..........48
4.4 Experiment..........49
4.4.1 Experimental Information..........49
4.4.2 Proposed Early Termination Algorithm Experimental Results..........49
Chapter 5..........Conclusion and Future Research 58
Reference..........60

[1]Moving Picture Expert Group (MPEG). (n.d.). Official Web Site. [Online] Available: http://mpeg.chiariglione.org/. Accessed July 18, 2019
[2]I. E. Richardson, “The H.264 Advanced Video Compression Standard 2nd ed,” Wiley, New York, 2010.
[3]Video Coding Expert Group (VCEG). (n.d.). Wikipedia. [Online] Available: https://en.wikipedia.org/wiki/Video_Coding_Experts_Group. Accessed July 18, 2019.
[4]G. J. Sullivan, J.-R. Ohm, W.-J. Han, and T. Wiegand, “Overview of the High Efficiency Video Coding (HEVC) Standard,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 22, no. 12, pp. 1649-1668, Jan. 2012.
[5]V. Sze, M. Budagavi, and G. J. Sullivan, “High Efficiency Video Coding (HEVC) – Algorithm and Architectures,” Springer International Publishing Switzerland, New York, 2014.
[6]x265. (n.d.). Official Web Site. [Online] Available: http://x265.org/. Accessed June 28, 2019
[7]H. Zhang, and Z. Ma, “Fast Mode Selection Algorithm for HEVC Intra Encoder.” IEEE International Conference on Automation, Electronics and Electrical Engineering (AUTEEE), pp. 169-172, May, 2013.
[8]L. Zhao, L. Zhang, S. Ma, and D. Zhao, “Fast Mode Decision Algorithm for Intra Prediction in HEVC.” Visual Communications and Image Processing (VCIP), Dec, 2011.
[9]Z. Y. Chen, S. Y. Chen, P. C. Chang, and J. S. Choi, “An Efficient Fast CU Depth and PU Mode Decision Algorithm for HEVC,” Proceedings of the 3rd International Conference on Multimedia Technology (ICMT 2013), pp. 153-163, Nov, 2013.
[10]K. Goswami, B.G. Kim, D. S. Jun, and J. S. Choi, “Texture Characteristic Based Fast Algorithm for CU Size Decision in HEVC Intra Coding,” 7th International Conference on Integrated Circuits, Design, and Verification (ICDV), pp. 88-93, Oct, 2017.
[11]T. Katayama, K. Kuroda, W. Shi, T. Song, and T. Shimamoto, “Low-Complexity Intra Coding Algorithm Based on Convolutional Neural Network for HEVC,” International Conference on Information and Computer Technologies (ICICT), pp. 115-118, May, 2018.
[12]X. Yu, Z. Liu, J. Liu, Y. Gao, D. Wang, “VLSI Friendly Fast CU/PU mode decision for HEVC Intra Encoding: Leveraging Convolution Neural Network.” IEEE International Conference on Image Processing (ICIP), pp. 1285-1289, Dec, 2015.
[13]Z. Liu, X. Yu, S. Chen, and D. Wang, “CNN Oriented Fast HEVC Intra CU Mode Decision,” IEEE International Symposium on Circuits and Systems (ISCAS), pp. 2270-2273, Aug, 2016.
[14]Y. Li, D. Liu, H. Li, L. Li, F. Wu, H. Zhang, and H. Yang, “Convolutional Neural Network-Based Block Up-Sampling for Intra Frame Coding,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 28, issue 9, pp. 2316-2330, July, 2017.
[15]Y. C. Lin, and J. C. Lai, “Edge Density Early Termination Algorithm for HEVC Coding Tree Block,” IEEE International Symposium on Computer, Consumer and Control (IS3C), Taichung, Taiwan, pp. 39-42, June, 2014.
[16]Study of Dendrimers in Cerebral Palsy Animal Model Points to New Therapeutic Target. Cerebral Palsy News Today, https://cerebralpalsynewstoday.com/2016/05/04/microglial-cells-interaction-with-dendrimer-in-cerebral-palsy-may-lead-to-novel-therapeutics/, May, 1998.
[17]K. Suehring, K. Sharman, (2015). HEVC Reference Software 16.7. [Online]. Available: https://hevc.hhi.fraunhofer.de/. Accessed June, 28, 2017.
[18]M. Mazur, “A Step by Step Backpropagation Example,” https://mattmazur.com/2015/03/17/a-step-by-step-backpropagation-example/, , Mar, 2015.
[19]Artificial neural network. (n.d.). Wikipedia. [Online] Available: https://en.wikipedia.org/wiki/Artificial_neural_network. Accessed July 18, 2019.
[20]Convolution neural network. (n.d.). Wikipedia. [Online] Available: https://en.wikipedia.org/wiki/Convolutional_neural_network. Accessed July 18, 2019.
[21]G. Bjontegaard, “Calculation of average PSNR differences between RD_curves,” 13th VCEG-M33 Meeting, Austin, Tx, April 2001.