隨著影像編碼技術發展，許多影像編解碼標準已廣泛應用於生活中， 如MPEG-4可用於網路串流影像、 H.264/AVC 應用於高畫質影像。在這些影像編解碼標準中應用了許多的編碼技巧來提高編碼效率，如畫面內預測(Intra Prediction)，離散餘弦轉換(Discrete Cosine Transform)以及多張參考畫面(Multi-Reference Frame)。然而，這些用來提高編碼效率的技術需要大量的運算資源及記憶體頻寬。隨著半導體製程技術的演進，大量的運算資源可透過適當的硬體設計技巧或軟硬體整合的方式來解決，但大量的運算資料仍須透過對外部記憶體作存取，如H.264中移動估計(motion estimation)就必須從外部記憶體中大量存取所要參考的資料來做運算，在記憶體頻寬或容量有限的影像編碼系統中會造成設計上的一個瓶頸。嵌入式壓縮(Embedded Compression)技術可預先將資料經壓縮後再存入外部記憶體中，藉此來減少對外部記憶體的使用量。 本論文所提出的嵌入式壓縮演算法主要包含下列技術: (1) Side-Oriented Prediction (2) Adjusted Binary Code (3) Two-Level Rate Control Scheme. 應用two-level rate control 的嵌入式壓縮演算法不僅可以有效率的控制壓縮率，並且可減少因量化而導致影像品質失真的情形。實驗結果顯示本論文所提出的演算法在失真度上與未加入嵌入式壓縮演算法之前相比只有1.33%的下降，且與預期的壓縮率比較也僅有1.60%的差距。 因此，在現今記憶體受限的影像編碼系統中，本演算法可提供一個有效降低記憶體頻寬及容量的方法。Owing to the rapid evolution of video coding techniques, video coding standards have been widely applied in many applications, such as MPEG-4 for streaming video on Internet and H.264/AVC for High Definition TV (HDTV). Many coding tools, including intra prediction, discrete cosine transform (DCT) and multi-reference frames (MRF) are further incorporated to improve coding efficiency. With these coding functions, great demands for computation resource and memory bandwidth are significantly increased. As the semiconductor have advanced into nano-scale technology, huge requirement of computing operation can be satisfied by proper design methodology, such as HW/SW co-design or dedicated hardware for acceleration. However, memory access operation still occupies a high percentage of whole computing profiling in H.264/AVC encoder. Therefore, memory bandwidth becomes a bottleneck in video coding systems. The embedded compression (EC) technology is widely applied to save frame memory size and bandwidth requirement. For video coding system, the proposed EC algorithm mainly consists of three core techniques: side-oriented prediction, adjusted binary code and two-level rate control scheme. With two-level rate control scheme, not only the compression ratio (CR) can be precisely controlled, but also the visual quality is maintained. Experiment results reveal that this work achieves the PSNR drop of 1.33%, and the error between CR and Target CR (TCR) is as minor as 1.60%. Consequently, this work is quite suitable for saving memory bandwidth in video coding system.
