计算机应用 ›› 2010, Vol. 30 ›› Issue (06): 1626-1628.

• 图形图像处理与模式识别 • 上一篇    下一篇

基于H.264的多参考帧运动估计快速算法

严浩1,何小海2,卿粼波1,吕瑞1,曾强宇1   

  1. 1. 四川大学电子信息学院图像信息研究所
    2. 四川大学图象信息研究所
  • 收稿日期:2009-12-03 修回日期:2010-01-26 发布日期:2010-06-01 出版日期:2010-06-01
  • 通讯作者: 严浩

Fast multi-frame motion estimation algorithm for H.264

  • Received:2009-12-03 Revised:2010-01-26 Online:2010-06-01 Published:2010-06-01

摘要: H.264/AVC中引入多参考帧运动补偿来提高视频编码性能,由此产生的多参考帧运动估计(MRF-ME)却带来了巨大的运算代价。为提高编码速度,降低计算复杂度,提出一种基于空间域相关性的运动估计算法——缩小的菱形算法(DDS)。先运用前向主矢量选择法不断修正预测运动矢量,再根据最佳参考帧位置的统计特性对不同参考帧使用不用模板进行搜索。实验结果表明,与H.264参考模型JM10.2相比,该算法保持了较好的图像质量且码率变化很小,运动搜索点数平均减少接近80%,并能有效地降低编码器复杂度。

关键词: 多参考帧, 运动估计, 空间域相关性, 菱形搜索

Abstract: The multiple reference frame motion compensation was introduced into H.264/AVC to improve video coding performance. However, the resulted Multiple Reference Frame Motion Estimation (MRF-ME) leads to much more computational cost. To improve the encoding speed and reduce the computational complexity, a new algorithm called DDS based on spatial correlation was proposed. First, the algorithm updated predictive motion vector constantly using Forward Dominant Vector Selection (FDVS), then searched different reference frame by different template according to the statistical characteristic of the location of optimal reference frame. The experimental results show that the algorithm can decrease the search points by 80% and reduce the complexity of the encoder considerably while maintaining nearly unchanged Peak Signal-to-Noise Ratio (PSNR) of pictures and the bit rate compared with the H.264 reference software JM10.2.

Key words: multiple reference frame, motion estimation, spatial correlation, diamond search