Fast algorithm and efficient hardware architecture of half-pixel interpolation unit for H.264/AVC

Abstract

A fast half-pixel motion estimation algorithm and its corresponding hardware architecture are presented. Unlike three steps are needed in typical half-pixel motion estimation algorithm, the presented algorithm needs only two steps to obtain all the interpolated pixels of an entire 8 × 8 block. The proposed architecture works in a parallel way and is simulated by Modelsim 6.5 SE, synthesized to the Xilinx Virtex4 XC4VLX15 Field Programmable Gate Array (FPGA) device, and verified by hardware platform. The implementation results show that this architecture can achieve 190 MHz and 11 clock cycles are reduced to complete the entire interpolation process in comparison with typical half-pixel interpolation, which meets the requirements of real-time application for very high defination videos.