H.264中有效的全零块预测方法

Efficient prediction method for all-zero blocks in H.264

在低比特率视频编码中,全零块预测是一种常用来优化编码器的技术。目前几乎所有的方法都是采用绝对误差和SAD)进行全零块预测,但在H.264中采用哈达玛(Hadamard)变换编码时,这种方法不能直接使用。采用高斯分布分析H.264视频编码中的残差系数,提出了一种基于变换绝对差值总和(SATD)预测全零块的有效方法。通过实验结果表明,该方法在保持图像质量的同时,提前预测出包括全零块在内的四种特殊编码块,减少了离散余弦变换(DCT),量化(Q)的运算量,提高了H.264的编码效率。

In low-bits rate video coding,All-Zero Blocks（AZBs） prediction is a common technique to reduce the computational complexity of the encoder.Previously, all the AZBs prediction approaches employ the Sum of Absolute Difference （SAD） available ahead of DCT and quantization for early detection.However,when the Hadamard transform is enabled for H.264 encoding,the method can not be used directly.The Gaussian distribution is applied to study the residual coefficients in H.264 encoding.The paper presents an efficient approach for detecting using the Sum of Absolute Transformed Difference （SATD）.It is shown by experimental results that the proposed method can early predict four kinds of special blocks including AZBs and achieve the best performance in reducing the DCT and Q computations and obtain almost the same video quality as the original encoder in H.264.