基于SVM的跳帧转码快速帧间编码算法

Fast inter coding in frame-skipping transcoding based on SVM

为降低H.264跳帧转码运算复杂度,在帧间模式选择中引入支持向量机(support vector machine,SVM)技术,提出一种SKIP模式提前判决算法,具体讨论了关键的特征向量与核函数的选择方法。同时,利用宏块时域相关性缩减预测模式数量,实现率失真优化提前终止。其次,通过定义宏块运动程度参数,针对不同运动程度的视频序列自适应地选择运动矢量合成算法和运动搜索策略,有效减少在已获得最佳运动矢量后仍进行运动搜索的运算冗余。实验证明,此算法极大降低了算法复杂度(平均84%)的同时保证了较高的视频质量,与现有算法相比,具有更强的鲁棒性和更广泛的适用性。

In order to reduce the computational complexity of H. 264 frame-skipping transcoding, an support vector machine （SVM） technology is employed in inter mode decision process, and a SKIP mode decision algorithm is proposed. Meanwhile, an early-termination strategy is used based on the temporal correlation between macro-blocks, and improbable candidate modes can be eliminated. In addition, an adaptive motion vector reconstruction and motion compensation algorithm is applied for sequences with different motions by measuring the motion activity of macroblocks, and the computational redundancy can be removed in case that the best matching motion vector can be found before motion search. Experimental results show that the coding time is reduced by an average of 84% dramatically while maintains the video image of high quality. Compared with the existing approaches, this algorithm is more robust and applicable.