Parallel Motion Compensation Interpolation in H.264/AVC Using Graphic Processing Units

Motion compensation interpolation is an important part in H.264 codec. With the growing interest on high-definition (HD) contents, it is important to find methods to reduce the time cost of motion compensation interpolation which has greatly hindered the implementation of real time coding for HD contents. In this paper, a forward step is developed towards the implementation of interpolation algorithm into a Graphic Processing Units (GPU) using Compute Unified Device Architecture (CUDA). The proposed algorithm has achieved high parallelism on GPU at pixel level. Experimental results show our proposed solution on GPU can be up to 193.61% over traditional method on Central Processing Unit (CPU) without introducing any distortions and leads to a CPU usage reduction at the same time, since the massive parallel processing capability of GPU is efficiently harnessed.