基于朴素贝叶斯分类的DVC-HEVC快速转码

Fast transcoding of DVC-HEVC based on naive Bayes classification

分布式视频编码(DVC)与传统视频编码之间的转码为移动终端设备之间的低功耗视频通信提供了一种有效的实现思路。以DVC与HEVC转码为研究对象,利用DVC解码端信息,针对高效视频编码(HEVC)中复杂度极高的编码单元(CU)划分过程进行复杂度优化研究。在DVC解码端提取与CU划分相关的纹理复杂度、运动矢量及预测残差3种特征信息;在HEVC编码端基于朴素贝叶斯原理建立CU快速划分模型,模型生成后便可以通过输入特征信息对当前CU划分进行快速决策,避免大量率失真(RD)代价计算过程。实验结果表明,本方案在编码比特率略有上升的情况下大幅缩短了HEVC编码时间,平均下降幅度达到58.26%,且几乎不影响视频质量。

The transcoding between Distributed Video Coding(DVC) and traditional video provides an effective way to realize low-power video communication between mobile terminal devices. In this paper, DVC and HEVC(High Efficiency Video Coding) transcoding is taken as the research object, and the decoding end information of DVC is utilized to carry out complexity optimization research on the division process of coding unit with high complexity in HEVC. Firstly, texture complexity, motion vector and prediction residuals related to Coding Unit(CU) partition are extracted from DVC decoder. Then, based on naive Bayes principle, a CU fast partition model is established at the HEVC coding end. After the model is generated, current CU division can be quickly determined by inputting feature information, avoiding a cost calculation process of large number of Rate Distortion(RD). Experimental results show that the proposed scheme significantly reduces the HEVC coding time by 58.26% with a slight increase in the encoding bit rate, and hardly affects the quality of video.