H.264视频编码器中运动估计算法的优化

Optimization of Motion Estimation Algorithm in H.264 Video Encoder

摘要: 运动估计是视频编码的关键技术，H.264编码器中的运动估计采用UMHexagonS算法，具有良好的综合性能，但其运算量仍比较大，难以满足实时编码的要求。本文基于MVFAST算法图像运动复杂度，提出了一种UMHexagonS的改进算法。依据ROS(支撑域)，通过在水平垂直的四个方向上分别设置搜索范围，建立了非搜索中心对称的动态窗口，用更小的搜索范围覆盖了最佳匹配点，进一步减小了搜索点数；进一步根据图像运动复杂度，优化了起始搜索模板的选择策略，直接减少了中等和低等运动复杂度视频序列的计算量。实验表明，与UMHexagonS算法相比，改进后的算法在获得几乎同等重建图像质量和码率的同时，运动估计时间平均节省了24.53%；与对照文献相比算法节省了6.24%，有效的提高了编码器的实时性。

Abstract: Structural Motion estimation is the key technology of Video coding, The UMHexagonS algorithm of H.264 encoder has a good overall performance, while its computation is still relatively large to meet the requirement of real-time encoding. In this paper, based on the complexity of image motion in MVFAST algorithm, an improved UMHexagonS algorithm is proposed. Based on ROS (region of support), by setting search range separately in the four directions of horizontal and vertical, a dynamic search window was established to covering the best match point using a smaller search window; Further according to the image motion complexity, the selection strategy of initial search template was optimized , which directly reduce the computation of low and middle complexity of motion video sequence. Experiments show that, compared to UMHexagonS algorithm, our optimized algorithm can save about 24.53% ME time but almost has no change in the reconstructed picture quality and bitrate; and ME time is 6.24% lesser than compared algorithm as well as enhance the real-time performance of the encoder.