臺灣博碩士論文加值系統

在這篇論文中，我們提出兩種應用於視訊壓縮中區塊移動預測的快速搜
尋演算法：叢聚搜尋演算法(CSA)和雙向搜尋演算法(2WS)。同時，我們也
提出一個新的區塊比對方法：動態區塊型態比對法(DSBM)。CSA以叢聚技
巧選擇每個區塊內最具代表性的像素，並再經由中心偏頗的概念精鍊最後
找出的移動向量。CSA 是一種以特徵像素為基礎的搜尋演算法，此法以減
少每個區塊中所需計算像素的數量來改進搜尋速度。2WS是一個以搜尋位
置為基礎的搜尋演算法，此法以減少前一個畫面中搜尋視窗內需檢查的區
塊個數來改進搜尋速度。2WS並非只簡單地於每個搜尋步驟中選擇兩個可
能的?

In this thesis, we propose two new fast search algorithms in
block motion estimation (BME) : the clustering search algorithm
(CSA) and the two-way search algorithm (2WS).Besides CSA and
2WS, we also propose a new block matching method, named the
dynamic-shaped block matching (DSBM), to achieve the better
image quality.Basically, CSA is a feature-base search algorithm.
It improves the search speed of BME by less calculated pixels
within each block. CSA selects the most representative pixels
of each block with the clustering technique, and refines the
final motion vector by the center-biased concept.2WS is a
position-base search algorithm. It improves the search speed
of BME by reducing the checked blocks of search window in
previous frame. 2WS does not only simply chooses two positions
in each step but also finds the motion direction between every
two steps. As for DSBM, it is a new block matching method whose
blocks in each frame of a video do not need to be square.
Moreover, all kinds of combinations of small blocks are taken
into consideration in DSBM.DSBM therefore provides a better
structure to get higher image quality in BME.