一种基于结构相似性的H．264／AVC运动预测方法

在H．264编码过程中，帧间预测的最佳匹配块的选择和编码模式的判决由率失真代价函数决定，在该函数中，通常对失真采用的衡量方法是绝对（变换）误差和（SATD）。相应地，其他类似的衡量方法，如MSE和PSNR也被用在质量评估当中。然而，SA（T）D和PSNR已经被证明不能反映人眼视觉对失真的真实的敏感程度。最近，人们提出了一种称为HSSIM（基于人眼视觉系统的结构相似性）的新的图像质量测度，由于更好地考虑了图像的结构信息，HSSIM与人类视觉的一致性优于PSNR和MSE。文中提出了一种用于帧间编码的新的运动预测方法（MEHSSIM），它建立在基于人眼视觉系统的结构相似性的基础之上。实验表明，在由HSSIM测量的主观视频质量基本保持不变的情况下，新方法可以使码率降低平均13．8％。

A Novel Motion Estimation Method Based on HVS-based Similarity for Inter Prediction in H.264/AVC

In H. 264, the optimal matching blocks and the prediction modes are determined by RDO cost function, in which, the distortion metric SA（T）D （Sum of Absolute （Transformed） Differences） is generally employed. Also, other metrics, such as MSE or PSNR are used for quality evaluation. However, it is preliminarily proved that SA（T）D and PSNR could not well reflect the real sensitivity of human vision to distortion. Recently, a new image quality metric named HSSIM （HVS based Structure Similarity） representing structural information is studied. Extensive experiments indicate that HSSIM could be more consistent with human vision sense than PSNR or MSE. In this paper, a new motion estimation method based on an improved structure similarity is proposed, that is, HVS-based structure similarity （MEHSSIM） for inter-frame encoding. It is proved by experiments that this new method could reduce an average the bit rate by 13.8% on the average while approximately maintain the comparable subjective video quality in the sense of HSSIM.