基於拉格朗日最佳化針對 H.264 畫面內編碼的位元率控制演算法

Abstract

為了讓視訊編碼後的位元率能維持在頻寬的限制之內，並且達到良好與穩定的畫面品質，位元率控制是相當重要的。然而目前大多數的研究都集中在畫面間編碼(inter frames)而不是容易造成緩衝區溢位的畫面內編碼(intra frames)。此外Intra-only 這種編碼方式也已經納入H.264 新的profile 當中，它較傳統GOP 編碼更適合應用在特別要求畫面品質的產品上。 在此論文中，我們提出一個於 Intra-only 編碼與GOP 編碼皆適用的位元率控制演算法。首先我們提出基於 Lagrangian 最佳化的 QP 決定方式，藉由位元率與 PSNR 預測模型，我們利用 Lagrangian 最佳化來找出可以平衡畫面品質與編碼效能的QP最佳解。另外，為了解決場景變換對位元率控制所產生的影響，在找出場景變換的畫面後，針對其使用以梯度複雜度為基礎的位元率模型來決定適合的QP，避免緩衝區溢位。實驗結果顯示，提出的方法可以達到更好更穩定的畫面品質，且緩衝內含量也都維持在較低的水平。

Rate control serves as an important technique to constrain the bit rate of video transmission over a limited bandwidth and to control the bit allocations within a video sequence to maximize its overall visual quality. However, most of rate control researches focus on inter coding frames instead of intra coding frames which are more possible to cause buffer overflow problem. Besides, H.264 Intra-only compression scheme has been standardized as H.264 profiles which are more proper for professional applications than traditional GOP compression scheme. In this thesis, we propose an improved rate control scheme which is appropriate not only for GOP compression but also for Intra-only compression. First, we present a Lagrangian-optimization based QP determination scheme for I-frames. By the estimation models for rate and PSNR of I-frames, the est quantization parameters can be determined by Lagrangian optimization method. In order to deal with the specific intra frames caused by scene transitions, a gradient complexity based QP determination method is proposed. After detecting scene change frames, the proposed gradient complexity based rate-QS model is adopted to determine appropriate QPs for avoiding buffer overflow and saving bit budget. Simulation results show, that compared to other reference algorithms, our approach achieves better and stable quality with low buffer fullness.