Abstract: A novel fast motion estimation algorithm is proposed in this paper. This algorithm reduces computation load by the full utilization of video statistical characteristics. The algorithm proposes one simple and efficient search pattern: the butterfly-shaped search pattern. By justifying the coefficients of DCT transform in advance, the algorithm can stop the search process in the half way. The median motion vector prediction technique is used for the selection of the initial search point in this algorithm. The result of simulation proves that this algorithm can achieve 43.26%~80% percent speedup compared with diamond search algorithm and the picture quality is better.

Key words: video image, motion estimation, block matching algorithm

摘要： 提出一种运动估计算法，该算法充分利用视频序列的统计特性以节省计算量，提高速度。并采用简单有效的搜索模板-蝶形模板，快速截止技术通过提前判断DCT变换系数全零来停止搜索。采用中值运动向量预测技术以确定初始搜索点。仿真表明，该算法与钻石搜索算法相比，搜索速度提高43.26%~80%，图像质量更好。

关键词: 视频图像, 运动估计, 块匹配算法

CLC Number: 

• TP301.6