【作者】 王兵；

【导师】 刘岩；

【作者基本信息】 哈尔滨工业大学 ， 计算机科学与技术， 2012， 硕士

【摘要】 转码(Transcoding)是一种将已经编码过的信号转换为另一种信号的技术，在视频上，视频转码的主要应用包括码率控制，帧率控制，分辨率调整以及视频格式转换等。然而，传统的视频转码方法多数只针对单视点视频。近年来，多视点视频凭借着更好的视觉体验受到广泛关注，也成为视频编码领域研究的热点。一般来说，多视点视频是使用位于不同位置的多个摄像机同步拍摄而成。因此，相对于传统单视点视频来说，多视点视频往往具有更大的数据量。为实现多视点视频在不同网络之间的传输和不同设备上的存储和播放，针对多视点视频的视频转码，尤其是减小码率的转码具有非常重要的意义。综上，本文提出了一种面向多视点视频降低码率的快速转码器。考虑到转码后的视频流选择的宏块模式和原视频的模式类似，原始视频中的宏块模式在转码时作为参考。同时，由于多视点视频相邻视点间的内容具有很强的相关性，所以相邻视点间相应宏块选择的模式同样具有相关性。在进行模式决策时，当前宏块在相邻视点内的相应宏块所选择的模式同样作为参考。这样，在转码时，只有部分模式需要考虑，取代了完全的模式决策，大大提高了转码效率。由于多视点视频在编码时使用了视点间参考，因此在转码时，不仅需要利用原始视频中同一视点内的编码信息，视点间的参考信息同样需要利用。参照传统视频转码中运动矢量修正算法，本文提出了视差矢量的修正算法。另外，由于转码前后相应宏块选择的宏块模式不一定相同，所以原始视频中的运动矢量和视差矢量不能直接利用。针对这种情况，本文提出了根据原始运动矢量和视差矢量，同时结合预测值来确定搜索中心的方法。通过该方法，可以把运动搜索限制在一个较小的范围内，在很大程度上提高了运动搜索的速度。另外，在多视点视频编码中，由于加入了视点间参考，对于一个宏块可选的参考帧数目大大增加。通过研究发现，在转码前后，所选择的参考帧也具有很强的相关性。因此，原始视频中的参考帧信息在转码时同样被利用。考虑到转码前后模式的变化，本文提出了一种“主导参考帧”选择的方法。在某些时候，比较费时的参考帧迭代搜索过程可以略去，转码效率进一步提高。实验表明，本文提出的多视点视频转码器具有较高的转码效率。与重新解码然后重新编码相比，在RD性能几乎不变的前提下，节省了超过90%的转码时间，具有很强的实用价值。更多还原

【Abstract】 Transcoding is defined as the conversion of one coded signal to another.Video transcoding is the operation of converting a video from one format intoanother format. A format is defined by such characteristics as the bit rate, frame rate,spatial resolution, coding syntax, and content. However, traditional transcodingmethod only applies to single view video. In recent years, multiview video isbecoming more and more popular because of better visual experience. Generalspeaking, multiview video is captured by different cameras on different positions atthe same time instance. Therefore, multiview video often has a larger amount of data.To transmit over various networks and store or play on different devices,transcoding for multiview video bit reduction is in badly need.We propose a fast transcoder for multiview video bit reduction. Considering themodes used in original video stream is very similar with that in transcoded video,the mode information extracted from original stream is utlized in our transcoder.Besides, the video content in adjacent views is almost the same so that the modes ofcorresponding MBs in adjacent views are also similar. Therefore, we utilized themode information of adjacent views. Then only a group of candidate modes areconsiderd for current MB when transcoding and much time is saved for modedecision.Since inter view reference is used in multiview video coding, the codinginformation for inter view reference shall be utilized when transcoding. Inspired bymotion vector (MV) refinement algorithm in H.264/AVC transcoding, we proposeddisparity (DV) refinement algorithm in proposed transcoder. The motion anddisparitu search center is calculated by the input MV/DV and the predicted one andthe search range is far smaller. Therefore, the most time consuming motion anddisparity search calculation in video coding is simplied.Besides, the number of reference frames in multiview video coding is morethan that in H.264/AVC. Through our simulation experiments, we find that there issome correlation in selected reference pictures before and after tranccoding. Theselected reference frames information in original stream is utilized in proposedtranscoder to accelerate the mode decision and search calculation. Considering themode used for transcoding may be different from that in input stream, we proposed adominant reference picture selection algorithm and sometimes the iterative searchfor bi-prediction can be omitted. The tranccoding efficiency is improved further.The simulation experiment results show that the proposed multiview videotranscoder saves over90%time compared to fully decode and re-encode transcoder while maintaining a high RD performance.更多还原