(from Chapter 1: Invited Addresses and Tutorials on Signals, Coding, Systems and Intelligent Techniques)
| Full Citation and Abstract
|
Title: |
Transcoding MPEG-2 compressed video into H.264/AVC |
|
Author(s): |
Daniele Bagni |
|
Address: |
Multimedia Coding SW Expert
Advanced System Technology labs
STMicroelectronics, Agrate Brianza, Italy
|
|
Reference: |
SSIP-SP1, 2005 pp. 35 - 35 |
|
Abstract/ Summary |
The emerging H.264/AVC standard (also known MPEG-4 part 10) achieves much higher compression at the same image quality than MPEG-2, MPEG-4 and H.263 and has a wider application spectrum in comparison with previous standards, ranging from broadcast HDTV transmission, to HD DVD, mobile phone and video streaming. For all these new applications, H.264/AVC will probably become the choice in the incoming years. Nevertheless, we have to take into account the huge amount of previously created, broadcasted or stored MPEG-2 material, during the last ten years of MPEG-2 'dominance' in digital video transmission at mid-high bitrates. To maintain compatibility we developed a 'transcoding' system, that is, a conversion process of transforming a compressed MPEG-2 video bitstream into an H.264/AVC bitstream, without necessarily doing a full MPEG-2 decoding step followed by a full H.264/AVC re-encoding one. The scheme of our transcoding is composed of two main stages: a complete MPEG-2 decoder and a 'light' H.264 encoder stage. We call it DBS, or Dynamic Bitstream Shaper, since it achieves syntax translation as well as bitrate compression. Our DBS works as a 'fast encoder', where all the most important H.264 coding decisions are inferred from the MPEG-2 input bitstream: during the MPEG-2 decoding phase we store and pass decoded parameters quantities like motion vectors, quantization factors and coding modes to the H.264 stage, thus avoiding a complete motion estimation process. In terms of performance, our experiments show a compression gain between 25% and 40% respect to the MPEG-2 input bitrate. We have also measured that our quality (PSNR) is at least equal or better that the full decode/re-encode ('trivial') solution. The implementation complexity of the DBS is anyway at least one order of magnitude lower than the full chain, due to re-using the information arriving from the input bitstream. Such a re-use allows us to skip the Motion Estimation process, which is by far the most ... |
|
|
We welcome your comments about this Article |