篇名 | 3D視訊編碼之運用多重視差向量於先進殘餘值預估方法 |
---|---|
卷期 | 158 |
並列篇名 | A Disparity Vector Derivation Method for Advanced Residual Prediction in 3D Video Codin |
作者 | 凃日昇 、 林敬傑 、 林俊隆 |
頁次 | 074-082 |
關鍵字 | 3D視訊壓縮 、 先進殘餘值預估 、 高效能視訊編 、 3D Video Coding;3DVC 、 Advanced Residual Prediction; ARP) 、 High Efficiency Video Coding;HEVC |
出刊日期 | 201408 |
隨著3D視訊需求的增加,國際視訊標準組織MPEG與ITU正合作制定3D視訊編碼標準3D-HEVC。此標準採納多項視訊編碼技術,其中有兩項技術分別為視訊間預測(Inter-View Prediction)及先進殘餘值預測(Advanced Residual Prediction; ARP)。前者利用拍攝不同視角的攝影機所擷取畫面之間的關聯性進行視差向量搜尋並產生預測殘餘值(residual),後者則是利用視差向量及動作向量(Motion Vector)找出參考視角(reference view)的殘餘值預估(residual predictor)。這篇論文提出一個改變ARP流程的方法,把原本ARP會從兩個權重值中挑選其一的流程,替換成用一個新的視差向量找出另一殘餘值預估。經由位元率-失真最佳化(Rate-Distortion Optimization; RDO)的計算評估後,編碼器會從原本的視差向量及新的視差向量兩者間選擇一個最佳的視差向量來完成ARP運算。這個方法不需要額外增加大量的運算,透過ARP現有編碼流程的改變,可以找尋到更佳的殘餘值預估,達到提高編碼效率之目的。我們將所提出的方法實作在標準組織所提供的參考軟體上,並遵照該組織所訂的測試樣本與條件進行實驗。結果顯示,此項技術對個別測試視訊的不同視角,能提升0.1%到0.7%以上的壓縮率。
Since the demand of 3D video coding technologies is emerging, two international standard committee MPEG and ITU worked together to develop a new 3D video coding standard 3D-HEVC. There are many coding tools in 3D-HEVC to enable the efficient coding of multiple camera views. Inter-View Prediction and Advanced Residual Prediction (ARP) are two important coding tools in 3D-HEVC. Inter-View Prediction firstly searches a reference block by a disparity vector and then gets residual data. ARP derives a residual predictor from the reference view by a disparity vector and a motion vector in the current view. This article presents an alternative disparity vector derivation method for ARP in 3D-HEVC. Instead of adaptively using two weighting factors in ARP process, the alternative disparity vector is applied to generate a new residual predictor. The encoder will choose the best disparity vector between the original disparity vector and the new disparity vector to perform ARP by rate-distortion optimization. Comparing to the current 3D-HEVC design, the proposed method improves the coding performance without extra effort. Experimental results show that the proposed method can achieve 0.1% to 0.7% BD-rate saving for the individual test sequence under the common test condition.