360度視訊以大量數據為代價提供觀看者身臨其境的體驗和豐富的資訊。對於 360 度視訊，立方體投影變體(例如:等角立方體貼圖(equi-angular cubemap,EAC))的多功能視訊編碼(Versatile Video Coding, VVC)，可實現比等矩形投影(equirectangular projection)更高的編碼增益。而VVC的參考軟體(VTM)採用基於R-λ模型之位元率控制(rate control)，其中存在著CTU層的SKIP區塊(SKIP block)與非SKIP區塊(non-SKIP block)混合著進行編碼，但R-λ模型卻不適用於SKIP區塊，導致R-λ模型的λ值無法準確計算，進而增加位元率誤差。因此，本論文提出以輕量化類神經網路(light-weight neural network)為基礎之SKIP 區塊預測器(SKIP block predictor)的畫面間編碼位元率控制演算法，若在位元率控制的位元分配階段後，當前CTU預測為SKIP CTU，則使用空間相鄰已編碼 CTU的λ值來計算當前CTU的λ值，使得位元率誤差降低。其中基於輕量化類神經網路之預測器的參數量非常少，其增加的運算負擔可被忽略。另外，本論文提出通過限制I畫面(intra frame)中的畫面層量化參數(quantization parameter)，以減少後續編碼畫面中的位元飢餓(starvation)。實驗結果顯示，本論文所提方案相較於VVC的參考軟體VTM-10.0的位元率控制方案，與現有可應用EAC格式的360度視訊之CTU層位元率控制方案，平均可達更低的位元誤差，且不會造成影像觀看品質降低。;360-degree videos provide immersive viewing experiences and rich information at the expense of huge amount of data. For 360-degree videos, Versatile Video Coding (VVC) of variants of cubemap projection (e.g., equiangular cubemap (EAC)) enables higher coding gains than equirectangular projection. The VVC reference software (VTM) adopts the R-λ model based rate control that is applied to both the CTU-level SKIP blocks and non-SKIP blocks. However, the R-λ model cannot work well for SKIP blocks so that the λ value of a SKIP block cannot be accurately estimated, thereby increasing the bitrate error. Therefore, this paper proposes a SKIP block predictor based rate control algorithm for VVC interframe coding, where the predictor is implemented using a light-weight neural network. After the bit allocation stage in the rate control algorithm, the λ value of the current CTU will be estimated using the λ value of a spatially adjacent coded CTU if the current CTU is predicted to be a SKIP CTU. Thus, the bit rate error is reduced. The light-weight neural network based predictor has a small number of parameters, and the increased computational load can be ignored. In addition, this paper proposes to reduce the bit starvation problem in later coded frames by constraining the frame-level quantization parameter in intra frames. Experimental results show that, compared with the original rate control scheme of the reference software of VVC (VTM-10.0) and the state-of-the art rate control scheme that can be applied to the EAC format 360-degree videos, the proposed scheme can achieve better bitrate accuracy without degrading the viewing quality of videos.
