任意影像形狀之JPEG 2000 離散小波轉換晶片設計

Abstract

離散小波轉換在許多的領域都扮演著重要的角色，例如訊源編碼、電腦視覺、語音識別和影像壓縮。因為小波轉換後的係數呈現能量集中於低頻的特性，很適合用在影像資料壓縮的前處理，許多影像壓縮的標準如JPEG 2000、MPEG-4都採用離散小波轉換，由於繁重的計算複雜度以及廣泛的應用範圍，有效率的離散小波轉換硬體架構也就顯得更為重要。 在這篇論文中，首先我們介紹離散小波轉換和提升式架構的基本理論，接下來我們改良提升式架構採用全管線化的硬體架構，因此我們可以操作在較高的工作頻率，並且我們合併了以四個編碼方塊為基礎的方式來使得JPEG 2000編碼系統具有更高的平行度，藉著改變離散小波轉換的輸出時序，內部的記體體需求可以減少為原本的四分之一，接著此架構也可以有效地支援可適應形狀的離散小波轉換，同時解決了邊界延伸和子取樣的問題，最後根據與其他架構比較的結果，來說明所提出之離散小波轉換架構的貢獻。

The discrete wavelet transform (DWT) plays an important role in many fields such as source coding, computer vision, speech recognition and image compression. Because the DWT coefficients have the property of energy conservation in the lowpass frequency band, it is suitable for image compression, such as JPEG 2000 and MPEG-4. With the heavy calculation and wide application domain, the research of efficient VLSI implementation for DWT becomes more significant. In this theory, we firstly introduce the basic theory of discrete wavelet transform and lifting scheme. Secondly, we adopt a fully pipelining VLSI architecture for lifting-based DWT. Hence, we can operate on higher working frequency. Besides, we merge a QCB (quad code block)-based DWT method to achieve the higher parallelism for JPEG 2000 encoding system. By changing the output timing of the DWT process, the internal tile memory size can be reduced by a factor of 4. Moreover, the architecture can efficiently perform shape-adaptive DWT and solve boundary extension and subsampling problems. Finally, according to the comparison results with other architectures, the efficiency of proposed architecture is proven.