基于可控金字塔的无参考图像质量评价模型

摘要

为具有多种失真的图像提供一个无参考质量评价通用模型,提出了基于可控金字塔的评价算法.该算法结合自然场景图像的统计特性,利用可控金字塔变换对图像进行多尺度多方向的分解,并经过较小的训练,得到未失真图像子带系数分布特征模型.根据不同子带系数对图像降质的影响设置权值,计算量化后失真图像系数的实际分布与未失真自然场景子带系数分布特征模型的偏离程度,以度量图像的质量.实验结果表明,该算法能较好地符合人类视觉的主观评价.%A novel universal model for no reference image quality assessment is proposed,which is based on the steerable pyramid algorithm.The algorithm builds the subband coefficients distributed model at multi-scale and multi-orientation about un-distortion images across small training,combining the statistics of natural scene features using steerable pyramid filter.The important weight value for subband coefficients is set according to the degree of influence on image degradation.Then,the image quality score can be computed through measure the deviate level between the distortion image coefficients statistics and the original subband coefficients distributed model.The experimental results show that the algorithm accord with human perception of quality.