Efficient integer motion estimation algorithm using sub-sampling
Sungjo Hwang, Jin Ho Ha, Myung Hoon Sunwoo · 2009
Motion estimation (ME) is an effective method to reduce temporal redundancy found in video sequences. Because the complexity of the full search block match algorithm (FSBMA) is extremely high, a number of fast algorithms has been proposed to reduce the computational complexity of FSBMA. This paper proposes a hardware-oriented integer ME (IME) algorithm based on sub-sampling. The proposed algorithm consists of two steps. First, global two point skip full search with 1/16 sub-sampling chooses five candidates. Second, local full search based on the five candidates is performed. Simulation results show that the proposed algorithm achieves almost similar coding performance as FSBMA in terms of PSNR and bit-rate. Moreover, compared with the exiting search algorithm, the proposed algorithms can reduce the sum of absolute differences (SAD) computational complexity by up to 36.7% and improve the PSNR by up to 0.05dB.