As demand for high-definition video services has increased in recent years, High Efficiency Video Coding (HEVC) was adopted as the next generation compression standard in 2013. Among newly added tools, Advanced Motion Vector Predictor (AMVP) technolog...
As demand for high-definition video services has increased in recent years, High Efficiency Video Coding (HEVC) was adopted as the next generation compression standard in 2013. Among newly added tools, Advanced Motion Vector Predictor (AMVP) technology helps improve compression performance by generating accurate prediction blocks in Motion Estimation (ME). However, when the ME is performed in parallel in a hardware based HEVC encoder, the AMVP information needs to be provided quickly for ME. For parallel processing of AMVP to speed up, there is a dependency problem in AMVP information generation. The size of AMVP hardware also increases sharply.
In this paper, two schemes are proposed for a fast and small AMVP generation module. Before ME, the Reduced AMVP scheme solves the data dependency problem while minimizing hardware resources by applying approximation. After ME, in the AMVP Reuse scheme, the approximation is corrected to follow the standard with small execution time. The proposed AMVP hardware is implemented and tested in Verilog. Simulation results show that the performance of the AMVP hardware is highly improved, while degradation in compression efficiency is negligible.