下一代的視訊編碼標準「高效率視訊編碼(High Efficiency Video Coding, HEVC)」,可以達到目前H.264/AVC標準的兩倍壓縮效能。在所有的編碼工具中,動作估計是減少畫面間冗餘資訊的核心技術。然而,為了要找到碼率-失真最小的匹配區塊,動作估計往往需要花費相當高的運算量。因此,本論文提出HEVC的一個快速整數點動作估計演算法。首先,我們觀察到目前HEVC的參考軟體HM 8.0,其動作向量預測器AMVP對於動作向量已有不錯的預測。我們以AMVP預測的動作向量為中心,依據碼率-失真選取三個搜尋方向;而這三個方向的搜尋將繼續延伸,直到無法找到碼率-失真更小的位置。我們也考慮以起始動作向量為中心進行十字樣式的搜尋機制。相較於HM8.0,本論文提出之演算法在整數像點的動作估計,可以節省平均約82 %的編碼時間(5倍快),而影像品質只有極些微的下滑。
The next-generation coding standard High Efficiency Video Coding (HEVC) can provide double coding efficiency as compared to the well-established H.264/AVC standard. Among all of the compression tools, motion estimation (ME) is a key element for eliminating the inter-frame redundancies. However, the involved computation in ME is high in order to locate the matching block with the minimal rate-distortion (RD) cost. In this paper, we present a new fast integer-pixel ME algorithm for HEVC. It is observed that the Advanced Motion Vector Prediction (AMVP) in the current HEVC reference software (HM 8.0) gives a good predictor for the initial motion vector (MV). The proposed method extends the search region from the initial MV to three directions that incur smaller RD cost. The directional search will be repeated until no further RD-cost reduction is achieved. A cross pattern on the AMVP predictor is also considered to increase the matching accuracy. As compared to HM 8.0, the proposed method saves averagely 82% time in integer-pixel ME (5 times faster) with very little rate-distortion degradation.