彩色图像质量评价的四元数梯度方法

Quaternion Gradient Method for Color Image Quality Assessment

多通道的图像质量评价方法是图像处理研究的一个难点.已有的研究表明四元数彩色图像表示为图像处理任务提供了有效的数学方法.梯度能够有效地刻画图像的结构特征,是构建图像质量评价指标的一类重要方法.针对彩色图像质量评价问题,首先将参考图像和失真图像用纯四元数表示,然后利用梯度掩模方法计算四元数梯度并设计梯度幅值相似性偏差方法.在彩色图像数据库TID2013和CSIQ上实验,结果表明了该方法优于目前流行的同类方法,计算效率高、与主观视觉感知具有较好的一致性.

The multi-channel image quality evaluation method is a difficult problem for image processing.Previous studies have shown that quaternion color image representation is an effective mathematical method for image processing tasks.Gradient is an important method to design image quality evaluation index based on its effective description for image structural distortions.This paper is firstly to represent both reference and distortion images with pure quaternion,then to compute gradient with mask method,and finally to design the similarity deviation method with gradient magnitude.Experimental results on TID2013 and CSIQ image databases show that the method is superior to some state-of-art gradient methods such as FSIM,GMSD,QSR-SIM.Also the proposed method has high efficiency and good consistency with the subjective visual perception.