標題: 即時動態影像預估器之設計
Motion Estimation Processor for Real-time Applications
作者: 何建宏
Jian-Hong Ho
溫 岸
Kuei-Ann Wen
電子研究所
關鍵字: 動態影像預估; 區塊對應搜尋法; 全區域搜尋; 三步搜尋; 平均絕對誤差; 管道; 記憶體模組化;;Motion;Block;Full;3-step hierachical;Absolute;Pipeline;memory modulization
公開日期: 1992
摘要: 用於動態影像壓縮上的動態影像預估,最常用的方式就是區塊對應搜尋法 。在本論文中,我們提出一個多功能影像預估器。使用者可根據他們在速 度或準確性的要求程度上選擇全區域搜尋法或三步搜尋法作影像預估。 文中的影像預估器可處理8x8點的區塊,搜尋範圍為7點。此預估器是 採用平均絕對誤差法做為比對的參考依據。 預估器的架構上,利用管道 方式來增加資料處理量,而且利用記憶體模組化及平行輸入方式,使得處 理資料可同時進行輸入及處理,如此便可省下資料輸入的時間。 由1.2 umCMOS的技術,因受限於記憶體存取速度,操作頻率可達10 MHz,而晶片面積為 0.9 cm x 0.9 cm。 In this thesis, a real-time motion estimation(ME) processor for videophone is proposed. The ME processor is multi-functional .It can perform Full Search (FS) and 3-Step Hierarchical Search (3SHS) to estimate the motion vectors. Users can select either FS or 3SHS under the constraint for speed or accuracy. The size of the the processing block is 8*8 pixels and the searching area is 7 pixels around the reference block. In the ME processor, we use the mean absolute error (MAE) as the matching criterion. To enhance the throughput,the pipeline strategy is used in the ME processor. The ME processor is constructed with memory part, register part, MAE processor, address generation part and control part. The ME processor operates image loading and vector estimation concurrently and hence speed-up the elapsed thime for the input of next estimated block. Under 1.2 um technology, the area of the motion estimation processor is 0.9 cm x 0.9 cm. The cored processing parts can achieve 20 MHz. Th match the SRAM derived from the labortory, the system is tuned down to 10 MHz. The simulation result shows that the processing speed is 10 MHz.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT810430100
http://hdl.handle.net/11536/56966
顯示於類別:畢業論文