Abstract
The H.265 (high efficiency video coding, HEVC) standard, approved April 2013, needs only half the bit rate of its predecessor, the H.264, which remains the most deployed video compression standard worldwide. H.265 can support screen resolutions ranging from 320 × 240 to 8K × 4K. The basic step of video compression is dividing a frame into plenty of coding units (CUs) and encoding in the zigzag scan order. Instead of H.264 macro-blocks, H.265 proposes a coding tree structure consisting of coding units (CUs), predict units (PUs) and transform units (TUs). For example, a 64x64 CU can be divided into four 32x32 CUs, a 32x32 CU can be divided into four 16x16 CUs and a16x16 CU can be divided into four 8x8 CUs. Furthermore, several new techniques are also proposed so as to further decrease the resulting bit-rate to double the coding efficiency than that of H.264. Apparently, the side effect of perceivable tremendous computational complexity should be solved by all means. In this thesis, cloud computing approaches to accelerate the speedup of the execution time of H.265 encoding are proposed with little bit-rate increasing and PSNR decreasing. Three parallel processing modes are suggested in H.265, naming slices, tiles, and wave-front. Both the slice-based and wave-front-based are considered and realized in parallel in this thesis. In the slice-based encoding, a frame is divided into several slices which are assuming independent to each other while encoding. Generally, load balancing is a challenge problem; a workload estimation and allocation method is proposed to balance the workload on the fly. As for the wave-front solution, it splits a frame into CU rows; dependences between rows should be synchronized and maintained carefully. To improve the efficiency of parallel processing, not only load balancing is a challenge, but the delays of synchronization must be as small as possible. Experiments conducted on 7-16 virtual machines system show that our implementations achieve better speedups, for slice-based, wave-front-based, respectively.