针对H.264采用多模式运动估计和模式决策等一系列新算法而引入的高计算复杂度问题,提出了一种基于预测的H.264编码快速模式决策算法.使用一种基于自适应阈值的提早退出检测算法,以较高识别率检测出静止宏块和区域运动宏块,及早作出模式决策;对未被判定的宏块使用基于预测的快速搜索算法得到所有4×4块的运动矢量;利用自底向上合并检测,得到其他模式块的运动矢量和最佳模式.实验结果表明,该算法与参考实现中的快速算法相比,仅损失了少量的图像质量和码率,但平均编码速度提高了30%.
H. 264 video coding standard has achieved significant improvement on coding efficiency due to the use of many new features, such as variable block-size motion estimation and mode decision algorithms. However, it also leads to much higher computational complexity. A novel fast predictive mode decision algorithm was presented to accelerate the processing of H. 264 encoder. An early-stop mechanism with adaptive threshold was proposed to determine the stationary and motion-followed macroblocks. A fast predictive motion vector estimation was performed on the 4 × 4 blocks in the macroblocks which were not earlydetermined. With these predictive motion vectors, a bottom-up merging method was performed to finish mode decision procedure. Experimental results show that the proposed algorithm can reduce the encoding computational complexity to about 70% of the reference fast algorithm, with negligible average peak signal-to-noise ratio dropping and small bi-rate increasing.