结合模式对应与机器学习的HEVC降分辨率转码

Resolution reduction transcoder based on machine learning and mode mapping for HEVC

HEVC是ITU-T VCEG继H.264之后所制定的新一代视频编码标准,它提高了视频的编码效率,在保证相同视频质量的前提下,压缩比与H.264相比提高了一倍。另外,随着4G网络的兴起和智能手机的普及,移动终端成为人们观看网上视频的一大主流平台。但是,网络中存储的视频分辨率普遍要大于移动终端的屏幕分辨率,为解决这个问题,开展了针对HEVC的降分辨率转码研究工作,利用高分辨率视频的编码信息,通过模式对应来简化低分辨率视频的编码模式的计算过程,并采用机器学习的方法来确定降分辨率时的组块阈值,以提高模式对应的准确性。实验结果表明,提出的算法与Trivial transcoder相比,在保持PSNR和比特率几乎不变的同时,编码时间平均节省了60%左右。

As the successor of H.264, HEVC is the newest video coding standard developed by ITU-T VCEG. It has obtained tremendous progress on video coding efficiency, compared with H.264, it can maintain the same video quality double the compression ratio. In addition, with the rise of 4G networks and the popularity of smart phones, mobile terminal has become a mainstream platform on which people watch online video. However the most of the video stored in the network has larger resolution than the mobile terminal. In consideration of this problem described above, in this paper the resolution reduction transcoder for HEVC is proposed. It reduces the computational complexity of the resolution reduction transcoder through mode mapping between the coding modes of the high resolution video and the low one, and machine learning is utilized to determine the chunking threshold so as to improve the accuracy of the mode mapping. Experimental results show that compared with the trivial transcoder with the bit rate and PSNR remain almost unchanged, the encoding time is saved by 63.51% averagely by the proposed method .