彈性規模視訊編碼之位元分配

Abstract

藉由利用彈性規模位元流, 信號傳輸之適應性與抗誤差性可被明顯的改善.而在以運動補償與分離式餘弦轉換為基礎的架構下實現多層傳送的主要缺點在於在相同空間與時間解析度下位元率的增加. 然而, 要達到之前所提到的改善, 必須付出代價. 我們研究使用分頻編碼而每個分頻以H.263+傳送去實現彈性規模編碼.而針對此主題我們只獲得了初步的結果. 我們的構想成果並未較H.263+定義的彈性規模編碼架構來得好. 然後, H.263+架構下的基本層與增強層間的最佳位元分配在此被研究. 多用戶傳輸系統與單用戶傳輸系統在本論文中皆被考慮, 針對這兩種情況分別以區塊集合與層為配置單位作位元配置. 首先, 我們發現Lagrange multiplier方法去作層位元配置在區塊集合位元配置的情況下是較為有效的, 而我們查表法較Lagrange multiplier方法更適合於層位元配置. 接下來, 區塊集合位元配置是為了更好的編碼彈性與成果而以略高的複雜被考慮的. 而即使區塊集合位元配置的複雜度並沒有高, 區塊集合位元配置仍舊不被採用, 因為無法正確預測增強層的位元率與失真的關係將導致其成果較層位元配置差.而層位元配置的模擬結果顯示我們的方法確實較眾人皆知的公共領域編碼為佳.

The flexibility and error resilience are significantly improved by employing a scalable bit stream. However, the major drawback of multi-layer representations within a motion compensated discrete-cosine-transform based framework is the bit rate increase as compared to a single-layered representation with the same spatial and temporal resolution as in the highest layer of the multi-layered representation. However, it is the price that need be paid to attain the aforementioned benefits. We study scalable coding based on subband coding with H.263+ for each subband. We have only obtained preliminary results for this subject. The particular construct that we employ does not yield benefits over the scalable coding framework defined in H.263+. The optimal bit allocation between base and enhancement layers under the H.263+ framework is then studied. Both the cases of multicasting system and single receiver system are considered. We do the frame-based bit allocation and GOB-based bit allocation f cases. We find that the Lagrange multiplier method is efficient in GOB-based bi allocation, while simple table look-up is efficient for frame-based bit allocation. The GOB-based bit allocation is considered for better bit allocation flexibility and better performance at slightly higher complexity. However, it is found to have the worse performance than frame-based bit allocation resulting from the prediction inaccuracy of rate-distortion relation of the enhancement layer. The simulation results show improvement in MSE of our technique compared to a well-known public-domain encoder.