JPEG2000之二维離散小波轉換資料路徑設計及具有內容感知之量化方法

Abstract

JPEG2000是新一代的JPEG影像壓縮標準，它擁有優越的效能並提供許多先進應用所需的特殊功能。為達到上述目的，JPEG2000導入許多最新發展的影像處理技術，而本論文著重於離散小波轉換(DWT)及其量化方法的探討。論文的第一部分首先比較了直接計算法、採用Winograd方式與採用提升式等三種不同的DWT演算法，並分析在不同影像尺寸與Motion JPEG2000影格速率所需要的運算複雜度；另外，我們提出採用分散式算術之低成本DWT實現方案。我們的數據顯示，在通用每秒三十張640×480畫面的壓縮規格下，此架構僅需要以乘加器為基礎之DWT架構45%的矽面積。本論文的第二部分探討了一些現存小波係數量化方法之效能，並提出具有內容感知的量化方法，其依據各個小波頻帶不同的能量來決定相對應的壓縮比，模擬結果顯示在同樣的資料壓縮率下，其可提供較好的PSNR。

JPEG2000 is the successor to the JPEG image compression standard, which offers superior compression performance and special features demanded by modern applications. It employs fundamentally different approaches and many recently developed techniques to achieve these goals. In this thesis, we have focused on the study of the cores in JPEG2000 – the discrete wavelet transform (DWT) and the embedded quantizer. In the first part, we compare the direct computation, Winograd algorithm, and lifting algorithm for DWT, and analyze the computational complexity under various application specifications of different image sizes and the so-called motion JPEG2000 conditions. Moreover, we propose a low-cost implementation based on the distributed arithmetic (DA). The implementation result shows that the DA-based lifting architecture requires only 45% silicon area of the MAC-based design for compression specification of common 30 640×480 frames per second. In the second part, we investigate the performance of existing quantization schemes for DWT coefficients. We then propose a content-aware quantization scheme, which allocates bits according to the power magnitudes of the DWT subbands. The simulation results show that it provides better PSNR than the conventional quantization schemes under the same bit rate condition.