視訊編碼泛運動參數估測之硬體架構

Abstract

在本論文中, 主要提出應用於混合式視訊編碼描述影像放大縮小及移動等 泛運動參數估測之硬體架構。由於連續畫面之相關性非常大, 於是這些運 動參數便可利用描述兩連續畫面之時空相關方程式所求得。利用區塊最小 均方根遞迴法找出運動參數的估計值, 使得區域性平均位移誤差達到最小 。利用本論文之硬體架構計算這些泛運動參數所須的時間能滿足即時需 求, 且計算之複雜度亦大大降低, 原因是本論文之硬體架構只用了很少的 乘法器。在本論文中, 將以傳統的方法與本文所提之泛運動參數估測作比 較, 並可發現使用泛運動參數估測確實使傳送之資料量大為減低, 而且效 果亦較傳統的方法為好。 An architecture for the computation of zooming and translational motion parameters for hybrid sequential video coding is proposed. These parameters are derived based on the spatial-temporal constraint between two consecutive frames. Block-recursive least mean square algorithm is used to estimate the parameters by minimizing the local mean squared displaced frame difference. The computational time is almost equal to the traditional block matching method, and computational complexity is significantly reduced by using only a few number of multipliers. Bits rate reduction with the global motion vector in video coding is analyzed. Performance comparsion with that of the conventional block matching method is made.