Merge prediction algorithm for adaptive parallel improvement of high efficiency video coding

The latest high efficiency video coding (HEVC) has increasing requirement to be implemented in a heterogeneous platforms or a GPU accelerators. However, due to the data dependency between neighboring blocks merge candidate lists of neighboring blocks cannot be generated in parallel and induce a bottleneck for parallel encoder designs. In this paper, a parallel merge mode prediction algorithm is presented, which removes the dependency of PUs within the merge estimation region and make the PU possible to do merge estimation in parallel. The obtained experimental results demonstrated that this implementation is with only 0.10% BD-rate loss, compared with the original reference software HM12.0.