H.264中多参考帧快速运动估计算法

针对H.264的运动估计计算量太大的问题,通过研究并验证视频多帧参考时的运动连续性,提出了一种基于有效区域的快速运动估计算法(VRF,Valid-Region-based Fast Motion Estimation). 该算法在第一个参考帧中用三步搜索(3SS,3-Step Search)快速估计整像素精度运动矢量, 并以此定义一个有效区域, 参考其它帧时, 在该有效区域内作改进的3SS估计; 然后选择最佳参考帧; 最后在所选择的最佳参考帧的有效区域内作全搜索和相应的分数像素精度估计. 实验证明, 和H.264全搜索相比, 本算法的运动估计搜索点数降低了82%以上, 而恢复质量(用峰值信噪比(PSNR,Peak Signal to Noise Ratio)表征)平均只下降0.24?dB,且码速率只增加8.81%; 和另一个经典的帧选择快速算法相比, 本算法的搜索点数降低了39%,且码速率平均下降了5.17%, 而恢复质量只下降0.08?dB.

Fast motion estimation algorithm for H.264 with multi-references

H.264, in spite of its high quality, is very time-consuming due to its complex motion estimation. A fast motion estimation (FME) algorithm called valid-region-based FME (VRF) was proposed by exploring the motion continuity property in multiple reference frames (RFs). VRF used 3-step search(3SS) FME in the first RF and restricts 3SS FME in other RFs to select the best RF. Due to motion continuity, it employed full search in the best RF only in a much smaller area, called valid region. Experiments were done to justify the motion continuity property. Results show that, compared with H.264 full search, VRF reduced the number of search pixels for motion estimation up to 82% while it decreased peak signal to noise ratio(PSNR)0.24?dB and increased bit-rate 8.81% averagely. Compared with another FME, VRF reduced the number of search pixels for motion estimation up to 39% and bit-rate 5.17%, with the PSNR decreased 0.08?dB averagely.