利用抽樣與門檻值來加速H.264禎間估計的快速演算法

Abstract

在H.264視訊壓縮演算法中用來消除冗餘資訊的方法主要有禎內預測 (intra prediction)跟禎間預測 (inter prediction)，其中禎間預測整體運算量超過百分之八十五，所以如果用預設的窮極搜尋法來做禎間預測的話，整體的效能會相當的差。在本論文中，我們提出一個藉由統計上的抽樣概念與門檻值方法來使禎間預測的速度提升並減少因抽樣方法造成的影像品質損失。在以往當我們要做禎間預測時我們必須計算每一個區塊上的每一個像素的能量值，然後才能找出最佳的配對；然而再加入抽樣方法後，我們藉由抽取部分的像素資訊做比對，而不使用窮極式的搜尋，如此一來便可以加快禎間預測的速度。另外一方面，為了提升預測的準確度，我們更進一步提出了門檻值方法。門檻值的做法採取的策略就是增加確認的區塊，當某個區塊抽樣值大於門檻值時，則此區塊就會被進一步的確認看是否為目前的最佳區塊。。在本論文的實驗中，所有的實驗模擬都是在JM17.2 [2] 上面所測試模擬的。

In this paper, a fast motion estimation algorithm applying to motion estimation by using sampling and threshold method is proposed for H.264. The algorithm adopts sampling method that only evaluates portions of pixels in a block to expedite the efficiency of motion estimation. There are three types of sampling modes those are illustrated in Section 3. On the other hand, in order to improve the accuracy of prediction, we further proposed a threshold mechanism. The objective of the mechanism is to increase the number of blocks to be investigated. Every pixel in a block will be estimated only when the sampling value of the block is less than the threshold value. The experimental results indicate that the proposed method has lower computational complexity than those of full search or fast full search of the H.264 reference software JM17.2. However, the method slightly sacrifices video quality, PSNR and bit rate.