3D視訊帶給觀看者強烈的立體感和臨場感，但其背後卻隱藏著龐大數據資料儲存與傳送問題，然而目前針對3D位元率控制(rate control)改良演算法，卻寥寥無幾，因此，本論文提出3D視訊編碼的色彩視訊之位元率控制演算法，共分為三個部分，第一部分為位元率控制之權重圖設計，由於位元率控制表現不佳，常為實際位元花費與分配位元不匹配，造成位元飢渴(bit starvation)，尤其在需使用高位元編碼區域更是嚴重，因此本論文針對畫面中高複雜度(high complexity)區域、同視角的畫面間之差異(inter-frame variation)區域、與同時間軸的視角間畫面之差異(inter-view variation)區域做結合定義為高花費位元區域。第二部分為位元分配最佳化(optimal bit allocation)，採用遞迴泰勒展開式(recursive Taylor expansion, RTE)，找出高花費位元區域與非高花費位元區域各所屬位元分配最佳解。此外由於編碼所採取之編碼方式為由上至下，由左至右，對於較晚編碼之編碼單元(coding unit, CU)，也極易發生位元飢渴，因此本論文之第三部分結合保留盈餘(retained earnings)概念於位元分配中，將對於合成影像品質重要區域，設計位元保留盈餘機制(mechanism of bits retained earnings)。由修改3D-HEVC(3D-High Efficiency Video Coding)之參考軟體HTM10.0之視訊編碼器實驗結果顯示，本論文所使用的演算法優於原始3D-HEVC參考軟體所採用之R-λ model的位元率控制，其數據顯示平均BDBR下降1.24%，平均BDPSNR上升0.035 dB，且平均位元誤差率為0.56%。;3D videos provide viewers strong stereoscopic feeling and telepresence, but it generates a huge amount of data with a hidden issue on data transmission and storage. However, there is few previous work focused on improving the rate control algorithms of 3D videos. Thus, this thesis proposes a rate control algorithm at the LCU level for 3D color video coding. This research is composed of three parts. The first part of this thesis is the weight map design for rate control. Since poor efficiency of rate control often occurs with the mismatch between the actual bitrate and the target bitrate, it easily causes bit starvation especially for regions which needs higher bits for coding. Therefore, this thesis defines a high bitrate cost region as the intra-frame high complexity region, inter-frame variation region, and inter-view variation region. The second part of this thesis is optimal bit allocation. We use recursive Taylor expansion (RTE) to find the optimal bit allocation for each region. In addition, because each frame is encoded from top to bottom and from left to right, the coding units (CU) that are coded later is more prone to bit starvation. In order to achieve better visual quality, the third part of this thesis proposes to adopt the novel concept of retained earnings for rate control. We design a mechanism of retained bits earnings to keep the image quality for the synthesized important regions. In our experiment, the implementation is based on 3D high efficiency video coding (3D-HEVC) software HTM10.0. The results show that our proposed performance is better than HTM10.0 with the R-λ model with 1.24% BDBR decrease, 0.035 dB BDPSNR increase, and 0.56% bitrate error.
