在多功能視訊編碼標準(Versatile Video Coding, VVC)的參考軟體中，對於採用畫面間編碼的編碼樹單元(coding tree units, CTUs)，其R-λ模型參數是參考自前一個位在相同畫面階層(frame level)的同位置(co-located) CTU編碼後更新的R-λ模型參數，然而，這種參數決策方法可能會造成部分CTUs的R-λ模型參數不準確，而影響位元率控制的效能。因此，本論文首先利用CTUs之間的R-λ曲線相似度，從相同畫面階層的前一個已編碼畫面中，選擇R-λ曲線與當前CTU最接近的CTU，作為理想的R-λ模型參數的參考CTU，並利用其編碼後更新的R-λ模型參數，進行CTU-level位元分配以及量化參數決策。經由實驗證實，目前VVC畫面間位元率控制的R-λ模型參數決策方法確實存在改善空間，若基於R-λ曲線相似度選擇R-λ模型參數的參考CTU，可進一步提升位元率控制的準確性與重建後的視訊品質。以上方法雖然可提升位元率控制的效能，但對於每個畫面間編碼CTUs，都必須經由大量的編碼才可獲得R-λ曲線，顯然不適合實際被採用。因此，本論文建立了CTU的特徵與R-λ曲線之間的關聯性，利用CTUs之間的運動資訊與紋理複雜度等特徵的差異性，選擇與當前CTU特徵差異性最小的CTU，作為R-λ模型參數的參考CTU。實驗結果顯示，本論文所提方案的平均位元率誤差約為0.57%，相比於VTM-14.0的位元率控制方案，平均BDBR下降了0.119%，平均BDPSNR提升了0.0012dB。;In Versatile Video Coding (VVC) reference software, the R-λ model parameters for coding tree units (CTUs) using inter-coding are referenced from the updated R-λ model parameters of previous co-located CTU at the same frame level. However, the parameter decision method may lead to inaccurate R-λ model parameters for some CTUs, and has an impact on the performance of rate control. Therefore, this thesis selects the CTU that has the R-λ curve is most similar to that of the current CTU from the previous coded picture at the same frame level first, and takes it as the ideal reference CTU of R-λ model parameters. Then, the updated R-λ model parameters of the ideal reference CTU are used for CTU-level bit allocation and quantization parameter decisions. Results demonstrate that there is still room for improvement in R-λ model parameter decision for VVC inter-frame rate control. The accuracy of rate control and reconstructed video quality can be further improved when the reference CTU of R-λ model parameters is selected based on the R-λ curve similarity. Although the above method can improve the performance of rate control, extra encoding time is required to obtain the R-λ curves for each inter-coding CTUs, which is obviously not suitable for practical use. As a result, this thesis establishes the correlation between CTUs’ features and R-λ curves, and uses the differences in the features of motion information and texture complexity among CTUs to select the CTU with the smallest feature difference from the current CTU, and takes it as the reference CTU of R-λ model parameters. Experimental results show that the average bitrate error of the proposed scheme is about 0.57%. Compared with the rate control scheme in VTM-14.0, the average BDBR of the proposed scheme is decreased by 0.119%, and the average BDPSNR is increased by 0.0012dB.
