| DC 欄位 |
值 |
語言 |
| DC.contributor | 通訊工程學系 | zh_TW |
| DC.creator | 陳伯豪 | zh_TW |
| DC.creator | Po-Hao Chen | en_US |
| dc.date.accessioned | 2023-1-16T07:39:07Z | |
| dc.date.available | 2023-1-16T07:39:07Z | |
| dc.date.issued | 2023 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:444/thesis/view_etd.asp?URN=109523034 | |
| dc.contributor.department | 通訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 5G 通信世代帶來高速且低延遲的高品質傳輸技術，此外隨著電腦運算速度加快， 視訊編碼標準從高效視訊編碼(HEVC/H.265)的 4K 畫質，提高到通用視訊編碼標準(Versatile Video Coding, VVC/H.266)的 8K 畫質。在 VVC 區塊編碼(coding unit, CU)架構下，除了在 HEVC/H.265 中的四分樹分割(Quad Tree, QT)模式，H.266/VVC 又增加二分樹分割(Binary Tree, BT)， 以及三分樹(Ternary Tree, TT)，即多分樹劃分(QuadTree plus Multi-Type tree, QTMT)。VVC畫面內預測模式也從原先的 35 種增加到 67 種。這些新技術讓 H.266/VVC 編碼效能大幅提升，但也需要更高的運算複雜度，實驗顯示H.266 畫面內編碼所需時間為 H.265/HEVC 的 18 倍。本論文針對 H.266 畫面內編碼架構，提出了結合卷積神經網路(Convolutional Neural Networks，CNN)與隨機森林分類器(Random Forest Classifier)快速編碼模式決策演算法。不同於原始VVC遞迴運算比較所有的切割模式，本論文一開始先利用卷積神經網路預測四分樹(QT)的切割與否，再透過隨機森林分類器來預測多類型樹(MT)的切割模式，藉此減少H.266/VVC 編碼中畫面內編碼的複雜度。最終實驗結果顯示，與VVC相比，整體平均BDBR上升1.48%的情況下，編碼時間大約可以節省68.63%。 | zh_TW |
| dc.description.abstract | The 5G communication generation brings high-speed and low-latency high-quality transmission technology. In addition, with the acceleration of computer computing speed, the video coding standard has been improved from the 4K quality of high-efficiency video coding (HEVC/H.265) to the universal video coding standard (Versatile Video Coding, VVC/H.266) 8K quality. Under the VVC block coding (coding unit, CU) architecture, in addition to the quad tree partition (Quad Tree, QT) mode in HEVC/H.265, H.266/VVC adds a binary tree partition (Binary Tree, BT ), and Ternary Tree (TT), that is, QuadTree plus Multi-Type tree (QTMT). The VVC intra-picture prediction modes have also increased from the original 35 to 67. These new technologies have greatly improved the encoding performance of H.266/VVC, but also require higher computational complexity. Experiments show that the time required for H.266 intra-screen encoding is 18 times that of H.265/HEVC. This paper proposes a fast coding mode decision algorithm combining Convolutional Neural Networks (CNN) and Random Forest Classifier (Random Forest Classifier) for the H.266 intra-picture coding architecture. Different from the original VVC recursive operation to compare all the cutting modes, this paper first uses the convolutional neural network to predict whether the quadrature tree (QT) is cut or not, and then uses the random forest classifier to predict the multi-type tree (MT) The cutting mode of H.266/VVC reduces the complexity of intra-picture coding in H.266/VVC coding. The final experimental results show that that the proposed fast encoding method can reduce up to 68.63% of encoding time with just 1.48% increase in BDBR as compared to the default VTM7.0. | en_US |
| DC.subject | 通用視頻編碼 | zh_TW |
| DC.subject | 畫面內預測 | zh_TW |
| DC.subject | 編碼單元 | zh_TW |
| DC.subject | 快速深度決策 | zh_TW |
| DC.subject | 隨機森林 | zh_TW |
| DC.subject | 卷積神經網路 | zh_TW |
| DC.subject | Versatile video coding | en_US |
| DC.subject | intra picture prediction | en_US |
| DC.subject | coding unit | en_US |
| DC.subject | fast depth decision | en_US |
| DC.subject | random forest | en_US |
| DC.subject | convolutional neural network | en_US |
| DC.title | 一種結合CNN與Random Forest 應用於H.266/VVC畫面內編碼之快速演算法 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Fast CU Partition for H.266/VVC Intra Prediction with CNN and Random Forest | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |