用於JPEG2000編碼器之資料傳送及緩衝器設計之最佳化

Abstract

在今日的靜態影像壓縮標準中，為了因應各種不同的需求，諸如運算能力、頻寬、有線及無線元件裡的記憶體等等，使得影像必須擁有可任意調整解析度和品質的功能。而JPEG2000就是被設計來符合這些需求的標準。它不但可以把這些需求放入一個獨立的位元串，也提供了許多以往標準所無法達成的特點。有鑒於資料的傳送和資料的儲存已經變成設計高速、低成本的影像處理架構裡最重要的關鍵，因此資料傳送及緩衝器設計的最佳化也就顯得更為重要。 在這篇論文中，我們提出了用於JPEG2000編碼器之資料傳送及緩衝器設計之最佳化。資料傳送的部分，我們結合了位元階級的zero run-length coding以及定址模式(addressing mode) 以降低離散小波轉換(DWT)和最佳斷點內嵌區塊編碼器(EBCOT)之間的傳輸次數，最多可降低到原本的35%。因此共用的系統匯流排可以盡量地保持空閒的狀態，使得整個系統的效能因此而得以提升。再者，資料傳送設計也可以幫助離散小波轉換和最佳斷點內嵌區塊編碼器之間的字元對位元平面轉換，並使得匯流排的資料寬度對影像品質和離散小波轉換的分解階層較不敏感。另一方面，藉著考慮上提式(Lifting-based) 離散小波轉換和最佳斷點內嵌區塊編碼器之間的編碼流程以及不同的設計參數，在離散小波轉換和字元對位元平面轉換所需使用的緩衝器最多可降低到原來的57%。

The different requirements of computing power, bandwidth and memory of wireless and wired devices, as well as emergence of diverse imaging applications, have made resolution scalability and quality scalability essential in today’s still image compression standards. JPEG2000 has been designed to support not only these requirements in a single bit stream but also a richest set of features that are not available by other standards. Due to the data transfer and storage have become the critical issues in designing high-speed and low-lost image-processing architectures, optimal data transfer and buffering schemes for JPEG2000 are essential. In this thesis, we propose the optimal data transfer and buffering schemes for JPEG2000 encoder. The data transfer scheme combines bit-level zero run-length coding and addressing mode to reduce data transfer times between Discrete Wavelet Transform (DWT) and Embedded Block Coding with Optimized Truncation (EBCOT) to 35%. Therefore the shared bus can be kept as free as possible, which in turn leads to increasing the potentiality of improvement on system performance. Furthermore, this data transfer scheme also helps in word-to-bitplane data format conversion from DWT to EBCOT and makes the width of data bus less sensitive to the quality of image and the decomposition level of DWT. On the other hand, by jointly considering coding flow of both lifting-based DWT and EBCOT with different design parameters, the reduction in total buffer requirement for both DWT and the word-to-bitplane conversion between DWT and EBCOT can up to 57%.