H.264/MPEG-4 AVC可調式視訊編碼之高資料效率預測演算法研究

Abstract

在設計可調式視訊編碼硬體的過程中，資料存取需求變成越來越困難，因為採用了需要密集作資料存取的層間(inter-layer)預測模式。為了解決可調式視訊編碼高資料存取量的問題進而達到編碼效能的提升。在這篇論文中，我們提出三個適用在層間預測模式上的高資料效率的演算法。首先，我們提出一個把畫面間(inter)預測和層間殘值(inter-layer residual)預測合併成一個預測程序的方法，並且重複使用資料來減少資料存取需求。其次，我們提出一個高資料效率的層間移動向量預測(inter-layer motion)演算法，藉由利用畫面間預測的參考資料來減少層間移動向量預測的大量的參考資料需求，還有提供一個適用在層間基底(inter-BL)預測的移動向量偵測機制，經由偵測層間基底的移動向量來決定參考資料是否可以被重複使用。此外，在移動估測議題方面，我們也提出一個根據畫面大小來適當調整的移動估計切換方法，藉由畫面的解析度大小適當地決定使用平行化之解析度移動估測設計加上層間高資料效率預測演算法或是畫面間及層間級別C(level C)的資料重複使用演算法。模擬結果顯示在影像大小為CIF時，傳輸資料率(bit-rate)僅有0.16%的增加，而影像品質(PSNR)只有0.002dB的下降; 在影像大小為480p時，傳輸資料率(bit-rate)則是0.91%的增加，而影像品質(PSNR)有0.09dB的下降。而以分析的結果來看，我們提出的方法可以在CIF以及480p的畫面大小下節省67.3%和80.9%的資料存取量。此資料存取節省量，不只可提高系統編碼效能，更可以達到記憶體存取功率消耗的減少。

Data access requirement becomes more and more critical in designing video coding hardware system especially for scalable video coding due to the adoption of inter-layer prediction with intensive data access requirement. In this thesis, we propose three data efficient prediction algorithms for inter-layer prediction to reduce the data access bandwidth. First, we introduce a method that combines inter and inter-layer residual predictions into one prediction process and reuse its data for lowering data access requirement. Second, a data efficient inter-layer motion prediction is presented by utilizing the reference data of inter prediction to further reduce the considerable reference data access. And a motion detection mechanism is proposed in inter-BL mode through perceiving the motion vector of inter-BL to decide that the reference data can be reused. Furthermore, for motion estimation issue, a frame size adaptive motion estimation switching method is also proposed in this thesis to adaptively determine the usage of the parallel multi-resolution motion estimation with data efficient inter-layer prediction algorithm or the level C data reuse scheme for both inter and inter-layer prediction algorithm by taking the frame resolution into consideration. Simulation results demonstrate that the average bit-rate increment is 0.16% and the PSNR degradation is 0.002dB in CIF sequences. For 480p sequences, the bit-rate increment is 0.91% and 0.09dB of PSNR degradation in average. However, analysis results show that the proposed method can save 67.3% and 80.9% data access requirement for CIF and 480p frame size, respectively. These data access saving not only can increase the coding speed of encoder but can save power consumptions brought by extreme data accesses for prediction.