多階層使用明可斯基不等式而得的動態估計影像壓縮

Abstract

在本論文中，我們根據兩種不同的評斷標準而提出了兩套新的快速整體搜 尋法，以便做動態影像壓縮裡的動態估計。我們的方法使用了金字塔架構 。由於我們已事先推導出該架構中每一層所得到的誤差測量值之間的單調 關係，所以許多不可能進入決賽的候選區塊能夠被及早丟棄，這使得我們 的方法能大量地降低計算時間。實驗結果顯示我們的方法所需的計算量確 實遠低於1995年刊登在IEEE期刊上的「連續消除法」所需。另一方面，正 如同「連續消除法」一樣，我們所提出的方法也可獲得和傳統「整體搜尋 法」相同的精確度。若與著名的「三步搜尋法」比較，雖然「三步搜尋法 」通常找到的並不是最佳解，其計算速度並不會比我們這種保證找到最佳 解的方法快很多。 In this thesis, two fast full search (FS) algorithms for motion estimation are presented according to two different matching criteria called Mean Absolute Difference (MAD) and Minimized Maximum Difference (MiniMax), respectively. Based on the monotonic relation between the distortion measures obtained for distinct layers of a pyramid structure, the proposed method successively rejects many impossible candidates considered in the FS and thus reduces the computation time significantly. Experimental results demonstrate that the computation load is much less than a recently introduced method known as the Successive Elimination Algorithm (SEA) of which the estimation accuracy is also the same as that of the FS. The algorithm proposed here has the property that the optimal result can be obtained with the processing speed not too far away from that of the Three-Step Search (TSS) which is one of the widely-used fast algorithms giving non-optimal estimation.