THESIS
2003
xiii, 77 leaves : ill. ; 30 cm
Abstract
The ISO MPEG-1 and MPEG-2 standards have made a major impact in home entertainment in recent years. The MPEG-4 standard encompasses advanced features and can achieve higher compression efficiency than MPEG-2 while keeping good visual quality. H.264 is recently being developed by ITU-T. A comparison between H.264 and MPEG-4 revealed that H.264 can achieve significantly better compression efficiency at the same visual quality. Some advance features of H.264 include 4x4 integer DCT, intra prediction in I frame coding, quarter-pixel motion compensation, multiple reference frames and multiple block size for P-frame coding, etc....[
Read more ]
The ISO MPEG-1 and MPEG-2 standards have made a major impact in home entertainment in recent years. The MPEG-4 standard encompasses advanced features and can achieve higher compression efficiency than MPEG-2 while keeping good visual quality. H.264 is recently being developed by ITU-T. A comparison between H.264 and MPEG-4 revealed that H.264 can achieve significantly better compression efficiency at the same visual quality. Some advance features of H.264 include 4x4 integer DCT, intra prediction in I frame coding, quarter-pixel motion compensation, multiple reference frames and multiple block size for P-frame coding, etc.
Intra prediction is important for I frame compression in H.264. Two block sizes of 4x4 and 16x16 are supported. There are 9 prediction modes for the 4x4 prediction, and 4 modes for the 16x16 prediction. All of the modes are designed using the pixels samples from neighboring blocks and based on some direction. In the reference software offered by the Joint Video Team (JVT), the full search (FS) algorithm is used to select the best prediction mode for a block. However, FS is also the computation bottleneck in I frame coding.
In this thesis, we have proposed an efficient intra prediction (EIP) algorithm. The computation is reduced by early termination, selective computation of highly probable modes, partial computation of the cost function and skipping the 16x16 intra prediction check. An adaptive threshold is exploited as well. EIP can significantly reduce the computational complexity of selecting the prediction mode, while keeping the similar PSNR and bit rate to the full search algorithm.
Post a Comment