A Fast Adaptive Statistical Genetic Motion Search Algorithm for H.264/AVC^1

In H.264, motion estimation algorithm plays a significant role in computational cost and coding efficiency. In this paper, we proposed a novel fast adaptive statistical genetic motion search algorithm for H.264/AVC.In our algorithm, we first make an early termination based on the estimation of stationary blocks by the predetermined thresholds with different block sizes. Based on the spatiotemporal predictive scheme, some starting points are well chosen to obtain good convergence rate. To adapt to fast motion and slow motion, a mutation mechanism is designed by exploiting the information of search range and direction based on the statistical distribution of motion vectors. Experimental data show that our proposed algorithm can achieve average speedup 29.3 with negligible average PSNR loss of 0.071 dB and average bit rate increase of 3.43% compared with full search algorithm. Compared with lightweight genetic search algorithm (LGSA) [9], our algorithm achieves speedup 7.92, PSNR gain 0.01 dB and bit rate decrease 14.96% averagely