H.264 is the most recent and efficient standard for video compression. This standard increases the compression ratio of its predecessors, by a factor of at least two, but at the cost of higher complexity. To reduce the encoding time, several H.264 encoders use a parallel approach. In this project, our primary objective is to design an approach that offers better acceleration than the parallel approach implemented by Intel in their H.264 encoder, delivered as sample code in the IPP library.
We present our multi-frame, multi-slice parallel video encoding approach and its motion estimation modes that offer various tradeoffs between acceleration and visual quality loss. The first mode, the fastest and the one that affects quality the most, performs motion estimation within the limits of the current slice. The second mode, slower than the first but offering better video quality, extends motion estimation to neighboring reference slices which have already been processed. The third mode, slower than the second, but providing even better quality, processes a slice only when all its reference slices have been processed.
Our experiments show that the first mode of our approach provides an average acceleration which is higher by about 55% than that obtained by the Intel approach. They also show that we obtain accelerations comparable to those obtained by state-of-the-art approaches, but without the disadvantage of having to use bidirectional frames. Furthermore, our approach can be implemented quickly in an H.264 encoder, which, like Intel’s, is based on a multislice approach.
| Date | 10 Jan 2011 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Stéphane Coulombe (Supervisor) |
|---|
Franche, J.-F. (Author),
Coulombe (Supervisor),
10 Jan 2011Student thesis: Master's thesis › Master in Engineering: Engineering