新興高效能視訊編碼技術

Abstract

本子計畫的主要目的是發展高效能視訊編碼(High Performance Video Coding, HVC) 的相關壓縮技術。特別是在高畫質(HD)與超高畫質(Ultra-HD)視訊的壓縮效能方面， 期望能夠獲致比使用傳統H.264/AVC 標準更好的壓縮效果。 近年來由於影像顯示與播放技術的發展快速成熟，現代人平日休閒活動的視覺饗宴 旋即被帶入高畫質(HD)影像的世界裡。早些年人們以為4Kx2K 解析度的影像，有如憑 欄欣賞窗外美景的氛圍已經是虛擬世界的極限，沒想到在8Kx4K 解析度的影像問世之 後，有如身歷其境般的感官衝擊讓許多人從此深深折服於超高畫質影像的魅力當中。 然而隨著影像的精確度和時間、空間、色彩的解析度逐漸提昇到下一個層次，連帶 的影像本身在傳輸過程中所消耗的資料位元率(Data Rate)也明顯增加，這讓現存的網 路系統在處理高畫質/超高畫質影像傳輸方面遭遇相當大的困境，因為過高的位元率， 造成在傳輸至終端使用者的過程中容易會有停頓的現象產生，大大限制了高畫質影像 的相關應用範圍。 因此，有鑑於高畫質影像的相關播放與壓縮設備即將普及，又有許多在壓縮效率方 面需要優先解決的相關事項，ISO/IEC Moving Picture Experts Group(MPEG)和ITU-T Video Coding Experts Group(VCEG)近期共同發表了Call-for-Proposal(CfP)競賽的 相關文件，希望能夠招來具高效能壓縮效率的新興技術。CfP 活動在產學界造成了相當 大地重視與迴響，一些有趣的初步成果顯示，兩倍以上於H.264/AVC 標準壓縮效率之 壓縮技術在下一代的標準實現將成為可能。 高效能視訊編碼標準目前仍處於草創階段，它帶來了全新的挑戰，也提供了研究者 一個絕佳機會發展新的技術，並推行視訊壓縮相關的尖端科技發展。在這個計畫中， 我們計畫修正基於方塊的混合式編碼架構(Block-Based Hybrid Coding Framework)， 在未來三年內發展出(1)空間多樣性之時間域預測方案(Spatially varying Temporal Prediction Scheme)，(2)高效率之動作取樣模型，和(3) 模式相依之差餘畫面編碼演 算法(Mode-Dependent Residual Coding Algorithm)。實際作法上我們首先將藉由稀 疏(Sparsely)、非固定性(Irregularly)的模式來針對動作向量取樣，接著盡可能使用 最佳化的方式重建時間域的預測器(Temporal Predictor)；第二步我們將針對動作取 樣的效率加以改善，使取得動作資訊(Motion Information)之訊號所造成的額外消耗 (Overhead)降到最低；最後一步我們將調整轉換(Transform)和差餘編碼(Residual Coding)以適用於具有空間多樣性之預測差餘(Prediction Residuals)。

This subproject aims at developing high-performance video coding (HVC) technologies that could potentially provide higher compression efficiency than the existing H.264/AVC standard especially in coding High Definition (HD) and Ultra High Definition (Ultra-HD) videos. The recent and rapid advances of the capture and display technology will soon bring to us a whole new experience of enjoying the already awesome HD programs. While the forthcoming 4Kx2K displays give the viewers a viewing experience reminiscent of sitting by a window, the experimental Ultra-HD systems capable of delivering images of size 8Kx4K will strike them with a stunning feeling as if they were actually at the site. As the fidelity and the spatial/temporal/color resolution of video signals are being taken to a next level, the video data rate also goes up explosively, making it difficult for the existing network infrastructure to deliver them economically to the end user. In anticipating that the HD- and Ultra-HD-enabled devices will soon be getting popular and affordable for consumer and professional use, the ISO/IEC Moving Picture Experts Group and the ITU-T Video Coding Experts Group have recently issued a joint (draft of) Call-for-Proposal to solicit HVC technologies. This call has generated tremendous research efforts in both academia and industry. Some interesting preliminary results indicate that a coding efficiency improvement by a factor of two over H.264/AVC may be achievable. The development of HVC is still in its early stage. It brings in new challenges but also provides opportunities for inventing new solutions and for pushing forward the frontiers of video coding technology. In this project, we plan to revisit the block-based hybrid coding framework. Efforts will be made over the next 3 years to develop (1) a spatially varying temporal prediction scheme, (2) an efficient motion-sampling pattern and (3) a mode-dependent residual coding algorithm. The first is to better reconstruct the temporal predictor based on sparsely and irregularly sampled motion vectors; the second is to improve the motion sampling efficiency so as to minimize the overhead necessary for signaling the motion information; and the last helps to adapt the transform and the residual coding for the spatially varying nature of the resulting prediction residuals.