Efficient Global Motion Estimation Using Macroblock Pair Vectors

An efficient MacroBlock Pair based GME technique (MBP-GME) is proposed. Firstly, we give the definition of macroblock pair and feature macroblocks, not feature points, are extracted from macroblock pairs with the principle of global motion vector synthesis. Secondly, simplified pixel-based gradient descent iteration is performed using the pixels in the feature macroblocks. Finally, the exact global motion parameters are computed after the iteration. Several test sequences are used to test the algorithm performance. Experimental results show that the proposed algorithm is accurate, fast and efficient.