https://scholars.lib.ntu.edu.tw/handle/123456789/122277
標題: | H.264 之轉換編碼及最佳化 Transcoding and Optimization of H.264 |
作者: | 陳宏銘 | 關鍵字: | H.264;MPEG-4;H.263;離散餘弦轉換;轉換編碼器;降採樣;多重方塊大小;移動估測;模式選擇;多重參考畫面;vH.264;DCT;transcoding;down-sampling;variable block-size;motion estimation;mode decision;multiple reference frames | 公開日期: | 31-七月-2005 | 出版社: | 臺北市:國立臺灣大學電信工程學研究所 | 摘要: | 本計畫以先前自行開發的研究成果為基礎,繼續探討MPEG-2至H.264之轉換編碼架構。此轉換 編碼器的功能是將視訊轉換至H.264格式,並將影像解析度降至原來的一半。在此架構中,我們將轉 換核心變換 (transform kernel conversion, 亦即從MPEG-2之離散餘弦轉換至H.264之近似離散餘弦)之 計算合併至降採樣(down sampling)的程序中。對Intra frame的轉換編碼,我們探討DCT-to-PEL和 DCT-to-DCT兩種轉換編碼架構。相形之下,DCT-to-PEL之運算較簡單。然而DCT-to-DCT的intra frame 轉換編碼尚未有人做過,我們是第一個完成理論性探討的研究單位。對inter frame的轉換編碼我們則 採用DCT-to-DCT之轉換編碼架構,盡量避開完全解碼再重新編碼的轉換編碼工作,減少運算量。 本計畫的另一目標為H.264最佳化,針對移動估測和模式選擇兩項關鍵模組,繼續研究簡化運算 量之方法。在移動估測方面發展可調變式搜尋策略 (包括動態移動向量預估器、提前終止,與動態搜 尋樣式) 與動態多重參考畫面選擇技術,將編碼速度增為原本參考軟體 (JM 8.5) 的六倍左右。在模式 選擇方面,利用方塊模式在空間與時間上的相關性,發展快速模式選擇演算法,將編碼速度提升約 兩倍左右。整合這兩種快速演算法於編碼器中,總編碼速度增加約十二倍,而編碼效能仍維持在原 來的程度,僅受非常輕微影響。 A key objective of this project is to develop a transcoder for converting an MPEG-2 video bitstream to the H.264 format. Along with the conversion, we also reduce the image resolution by half in each dimension. For computational efficiency, we avoid full decompression and full recompression in the spatial domain. The output bitstream is fully compliant to the H.264 standard. In our technique, the "transform conversion" between the DCT of MPEG-2 and the modified DCT of H.264 is incorporated in the down-sampling process. Inter and intra frames are processed separately. For transcoding of intra frames, both DCT-to-PEL and DCT-to-DCT architectures are considered, while for inter frames a DCT-to-DCT architecture is adopted. Another objective of this project is to improve the computational efficiency of two core modules, motion estimation and mode decision, of H.264. For motion estimation, we develop a fast algorithm with adaptive search strategy and intelligent reference frame selection. Compared with the H.264 reference software JM 8.5, this algorithm achieves, on the average, a 600% reduction of encoding time. For mode decision, we exploit the correlation between neighboring image blocks to predict the best possible mode. Compared with the reference software, this method achieves about 204% reduction of encoding time with negligible PSNR drop and bit rate increase. Integrating these two fast algorithms in the H.264 reference software, this combined method, on the average, is about 12.1 times faster in the total encoding time, at the cost of a slight PSNR drop. |
URI: | http://ntur.lib.ntu.edu.tw//handle/246246/20326 | 其他識別: | 932213E002137 | Rights: | 國立臺灣大學電信工程學研究所 |
顯示於: | 電信工程學研究所 |
檔案 | 描述 | 大小 | 格式 | |
---|---|---|---|---|
932213E002137.pdf | 698.27 kB | Adobe PDF | 檢視/開啟 |
在 IR 系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。