在現今異質無線網路與使用者設備相異的環境下，multi-view video與video plus depth等3D視訊格式皆不足以有效支援進階3D視訊應用，例如較廣範圍的 multi-view裸眼式立體顯示器，與自由視角視訊(free viewpoint video)，而 multi-view video plus depth (MVD)則較能克服multi-view video與video plus depth所面臨的問題。目前國際間3D視訊編碼相關研究大多著重於探討 multi-view video及video plus depth，MVD編碼技術仍少，相較於multi-view video，MVD雖包含較少視角視訊，但須額外傳輸深度視訊，開發MVD的texture video與depth video的內容相關性雖可提高壓縮率，卻將因此提升MVD編碼複雜度，因此，本子計畫研究目標為動態考量視訊內容統計特性，以降低模式決策(mode decision) ， 運動估測(motion estimation) ， 及視差估測(disparity estimation)計算複雜度為目標，並使解壓縮後之texture video 與depth video皆維持良好的視覺品質。 Either multi-view video or video plus depth is inefficient to support advanced 3D video applications in case of heterogeneous networks and user devices. On the other hand, although multi-view video plus depth (MVD) can support requirements of advanced 3D video applications such as autostereoscopic display and free viewpoint video in case of heterogeneous networks and user devices, it includes additional depth videos. Most current 3D video coding schemes are designed for the 3D video formats, multi-view video and video plus depth. Video codings of MVD have been rarely studied so far. Since exploiting the correlation of texture video and depth video to improve coding efficiency may increase computational complexity, this project will focus on reducing the complexity of mode decision, motion estimation, and disparity estimation based on on-line statistical analysis of video contents. With this, the visual quality of the reconstructed texture video and depth video can be retained without degradation of RD performance. 研究期間：10008 ~ 10107
