改进的空间下采样视频转码快速运动估计算法

摘要

提出一种基于输入码流信息和已转码码流信息的视频转码快速运动估计算法。本算法利用Alpha-激励均值滤波通过输入码流的运动矢量合成作为备选预测运动矢量之一，并利用H．264标准中帧间预测的方法通过已转码码流信息合成另一个备选预测运动矢量，共同作为EPZS运动估计算法的预测中心。结合图像的运动活跃性．自适应的调整运动估计的搜索策略。该算法比传统的运动矢量方法相比提高了1dB左右的PSNR，保持了较高的转码图像质量并与菱形搜索算法相比能够降低18％左右的转码时间。%An improved algorithm of Fast ME for down-scaling video transcoding is proposed based on inputted video information & transcoded video information. One of the predicted motion vector is synthesized by using the motion vectors and coefficients of the inputted stream with the Alpha-Trimmed Mean Filter, the other one predicted motion vector is synthesized by using the near macroblock of the transcoded stream with inter prediction, as the motion vector predictions. The search strategy of motion estimation is adjusted according to the predicted motion vector, and the frame activity.The algorithm produces about ldB PSNR higher than the traditional mv-composition algorithm. While preserving high transcoding quality, it reduces above 18% of encoder time as compared to the diamond search algorithm.