JPEG在過去被廣為使用的靜態影像壓縮系統,因為他在真實影像的壓縮處理上可以達到很高的壓縮比,而且運算複雜度也很低。然而,JPEG在進行低位元率(Low Bit-rate)的壓縮時,將產生令人難以忍受的方塊效應。JPEG2000 [1-5]是ISO/IEC JTC1/SC29/WG1所制定的最新靜態影像壓縮標準。JPEG2000在做低位元率壓縮時能提供比JPEG更平滑的壓縮品質,除此之外,它還加入了其他很多的功能,像是品質與解析度漸進式(Progrssive)的影像傳輸、興趣範圍(Region of Interest, ROI)編碼,同時支援無失真(Loseless)和失真(Lossy)壓縮,並且還有不錯的錯誤回復能力(Error Resilience)。因為JPEG2000有擁這麼豐富的特色,使得它可以使用在各式各樣的應用上,像是網路傳輸、數位相機、監視系統、數位電影院等。然而,也因為JPEG2000提供如此高的壓縮品質和這麼多的功能,所以它演算法的複雜度也遠高於JPEG。 記憶體的使用是JPEG2000晶片設計中最重要的議題。這篇論文提出一個沒有編碼區塊(Code-block, CB)記憶體的JPEG2000 Encoder電路架構。提出的架構藉由將2D-DWT(2D-Discrete Wavelet Transform)與EBCOT(Embedded Block Coding with Optimized Truncation, EBCOT)編碼順序(Scan Order) 一致化,來完全消除CB(Code-block)記憶體。在沒有CB記憶體的情況下,所提出的自適應方塊編碼(Adaptive Embedded Block Coding, AEBC)電路仍能夠跳過所有多餘的不重要位元平面(Insignificant Bit-plane, IBP)來進行編碼。節省它在處理IBP所花費的時間與功率。並以動態RDO(Dynamic Rate Distortion Optimization)來減少失真壓縮(Lossy Compression)時,EBC的運算量。除此之外,提出的架構所使用的Code-block-based DWT可以支援任意Tile大小與任意階層的DWT。整個所提出的JPEG2000 Encoder僅使用2.2KB的內部記憶體與1.5B/Cycs的外部記憶體頻寬,遠低於目前現有的其他架構。
The amount of memory required for code-block is one of the most important issue in JPEG2000 encoder chip implementation. To overcome the drawbacks caused by the large amount of code-block memory in JPEG2000, this paper proposes a new JPEG2000 encoder architecture without code-block memory. Here we try to unify the output scanning order of the 2D-DWT (discrete wavelet transform) and the processing scanning of the EBCOT (embedded block coding with optimized truncation) and further the code-block memory can be completely eliminated. Since the code-block memory has been eliminated, we propose another approach for embedded block coding (EBC), code-block switch adaptive embedded block coding (CS-AEBC) that can skip the insignificant bit-planes (IBP) to reduce the computation time and save power consumption. Besides, a new rate distortion optimization (RDO) approach is proposed to reduce the computation time when the EBC processes lossy compression operation. The DWT used in this work is a code-block-based DWT, and it can process any tile size of picture and any levels of DWT operation. The total memory required for the proposed JPEG2000 is only 2.2KB internal memory, and the bandwidth required for the external memory is 2.1B/cycle. Compared to other JPEG2000 architectures, our new approach has the cost and performance advantage.