具平型化處理之JPEG2000方塊編解碼晶片設計

Abstract

JPEG2000 是一種新的靜態影像壓縮規格，它擁有比JPEG更好的壓縮率，並也提供了更多的特色，但是相對的，JPEG2000也比JPEG需要更多的memory以及運算量，其中又以EBCOT為最，因此我們針對EBCOT裡的context formation提出了一些加快運算以及減少memory的方法。Sample-Skipping method可以直接對需要編碼的sample執行動作，略過不需被編碼的sample，而Pass-Parallel method可以使三個coding pass在同一層bit-plane上平行處理，column-based architecture則可同時判斷同一行的四個sample中是否需要被編碼，這三種方法可以有效的加速JPEG2000的編解碼速度，大約可以將編解碼所需的時間減少至36%。我們的設計經過CMOS 0.25製程合成後，晶片面積大小為1775 m × 1695 m.，工作頻率最快可以到達133 MHz，在100 MHz下，處理一張2304 × 1728的灰階影像時，編碼時間為0.323秒，解碼則需0.512秒。

JPEG2000 is a new still image compression standard. It has better compression performance than the JPEG standard and also provides new features not available in JPEG. However, the high performance and new features require more complex computations and hardware cost than traditional JPEG. Moreover, most of the computation time is in EBCOT. Therefore, an efficient JPEG2000 codec design is proposed to ease in the overhead. We focus on context formation module of EBCOT Tier-1 in JPEG2000. Two speedup methods, Sample-Skipping and Pass-Parallel, are adopted in our design. The Sample-Skipping method is to skip no-operation samples in each column and then codes the need-to-be-coded samples directly. The Pass-Parallel method is to process three coding passes of the same bit-plane in parallel to improve the system performance. A column-based architecture using these combined speedup methods is then proposed to check four samples in a column concurrently. The prototype chip of the proposed technique is synthesized in CMOS 0.25 m 1P5M technology. The area of this chip is 1775 m× 1695 m. The clock frequency can reach 133 MHz. With clock frequency, 100 MHz, it needs 0.323 second to encode and 0.512 second to decode an image with 2304 × 1728 image size.