Low complexity hardware implementation of reciprocal fractional motion estimation for H.264/AVC in mobile applications

Abstract

Motion Estimation, one of the most effective modules in H.264/AVC, constitutes 60%-90% of encoding time and computation. Close to %45 of this computation belongs to Fractional Motion Estimation (FME) which has to perform a time consuming half-pixel and quarter-pixel interpolation. In addition, interpolation is a major challenge for hardware implementation in real time, specially in mobile applications where processing and battery power is limited. Several modified sub-pixel accuracy search methods have been proposed in the literature in order to reduce the complexity of interpolation. The reciprocal method has been proposed to reduce the CPU encoding time for a software implementation of an H.264 encoder on personal computers. In this paper, the hardware implementation of the reciprocal method is evaluated and its PSNR performance and hardware cost are compared to that of a simplified RateDistortion Optimization (RDO) process. Simulation results indicate that using reciprocal FME with enabled RDO has less computational cost and better PSNR performance than using the conventional FME with disabled RDO.