空間時間斜率法在影像編碼上之應用

Abstract

就影像序列處理而言，動態估測(Motion estimation) 可說是使得一維及 二維信號的處理技巧得以有效運用到三維的影像序列處理的關鍵。傳統上 應用最普遍的動態估測法是全搜尋區塊比對法( Full search block matching method)。在本文中，我們嘗試以一個更能廣泛描述物體運動的 模式結合空間時間斜率法(Spatial-tem poral gradient method)作影像 序列壓縮之動態估測。依照物體運動方式之不同，在空間時間斜率法中又 可將運動參數的描述方式分為平移模式(translation model)及Affine模 式(Affine model)。由於是以最小平方法(least square method)解出估 測參數， 運算複雜度為 ，而全搜尋區塊比對法則是 。區塊大小是 N*N。我們模擬 MPEG 的環境，結合離散餘弦轉換 (DCT)，量化( Quantizat ion)和霍夫曼編碼法(Huffman coding)，比較空間時間斜率法 和全搜尋區塊比對法在影像壓縮編碼上的應用效能。實驗結果顯示，空間 時間斜率法將可獲得較低的傳輸位元率(Transmission bit rate )。 Motion estimation is the major concept for applying 1- D and 2-D signal processing techniques into 3-D image sequ ence processing.Traditionally,the full search block matchi ng method is the most widely used motion estimation method .In this paper, the spatial-temporal gradient method is con sidered for the motion estimation in image coding. Accordi ng to different motion behaviors,the motion parameters are described by translation model and Affine model. Because t he motion parameters are solved by the least square method ,the computational complexity is ,while the full search block matching method is ,where the block size is N*N. We simulate the MPEG environments combining DCT, quantizat ion, and Huffman coding to compare the performance of the spatial-temporal gradient method and the full search block matching method. Simulation results show that the spatial- temporal gradient method has a lower bit rate.