基于扩展梯度相似度的自适应图像质量评价算法研究

客观图像质量评价使用特定的数学模型使图像评价结果与主观评价一致。提出一种自适应的扩展梯度图像质量评价算法(AE-GSSIM),使用扩展梯度算子从多个方向计算图像梯度幅值,然后在梯度图像上定义并计算梯度相似度;考虑到亮度失真对图像质量的影响,为梯度相似度与亮度相似度设置不同的权重,使亮度相似度随梯度相似度自适应变化;最后,采用算术平均池化策略计算图像评价结果。在4个公开图像库中进行的实验表明,AE-GSSIM算法优于传统的评价算法,更加符合人眼视觉感知结果。

In order to evaluate the quality of an image more accurately,this paper proposes a novel objective image quality assessment algorithm-Adaptive Extended Gradient-based Structural Similarity(AE-GSSIM).Firstly,the extended gradient operator is employed to calculate the gradient amplitude from multiple directions;then,the gradient similarity is defined and calculated on the gradient image.Considering the effect of luminance distortion on image quality,different weights are set for gradient similarity and luminance similarity to make luminance similarity change adaptively with gradient similarity.Finally,the arithmetic average pooling strategy is used to calculate the image evaluation results.Experiments in four public image databases show that the AE-GSSIM algorithm is better than the traditional evaluation algorithm and is more consistent with the results of human visual perception.