Optimizing an H.264 video encoder for real-time HD-video encoding

Abstract: With the increased demands for higher resolution and higher quality video the requirements for larger storage medium and higher bandwidth has increased as well. One method to cope with these new demands is by introducing new ways to efficiently compress video. One problem with this approach is that better compression means higher computational complexity. This Master’s thesis presents three methods that are used to independently optimize an existing video encoder (using the H.264 codec). Where previous research has mostly focused on standard and lower resolution video, this thesis focuses on encoding HD-video (High-Definition). The implemented optimizations work differently, as an example some approaches makes better use of the computer hardware. The result of this work is that real-time encoding of HD-video, on a workstation computer, at minimum of 50 frames per second is achieved when combining instruction- and thread-level parallelism. Since the minimum requirement of real-time encoding is 25-frames it was also investigated how the extra complexity can be used to get better compression results.