基于结构相似度的快速运动估计算法

Fast motion estimation method based on structural similarity

基于人眼视觉系统和视觉域抽取的结构信息高度相关的原理,提出了一种基于结构相似度的快速运动估计算法(FMEBS)。该算法针对H.264率失真优化算法存在的不足,引入基于结构相似度的图像质量衡量标准,对失真度的表示进行修正,并采用快速的模式选择算法和有效的搜索模板。实验表明,在获得相近重建图像质量的前提下,FMEBS算法较之全搜索算法可节省约2.7%的比特率和91.2%用于运动估计的时间,较之UM-HexagonS算法可节省约1.9%的比特率和35.6%的时间。

Based on the assumption that the human visual system was highly adapted to extract structural information from the viewing field, this paper proposed a fast motion estimation method based on structural similarity（FMEBS）. Aiming at the existent defects of rate distortion optimized （RDO） for H. 264 inter coding, FMEBS introduced an image measurement called structural similarity（SSIM） to modify the distortion metric and used the fast inter mode decision approach and efficient search patterns. Experiments show that FMEBS reduces average 2.7% bit rate and 91.2% motion estimating time compared with FS and 1.9% bit rate and 35.6% motion estimating time compared with UMHexagonS while maintaining the similar reconstruction image quality .