基于多特征相似度的图像质量评价算法

Image Quality Assessment Algorithm Based on Multi-Feature Similarity

为了解决数字图像处理中的客观质量评价问题,本文提出一种结合梯度、相位以及边缘亮度特征融合的图像质量评价算法。该算法综合利用图像的梯度相似性、相位一致性以及边缘亮度相似性,较好地描述了图像的局部结构特征;同时,利用图像的梯度相似性和边缘亮度相似性,可以较好地体现出图像的对比度变化和显著性变化。通过多个特征相似度的融合,使得提出的客观评价方法与人眼视觉感知具有较好的一致性。通过在TID2008图像数据库的实验结果表明,对不同类型的图像失真,本文方法都具有较好的评价性能。

To address the problem of image objective quality assessment,an image quality assessment method is presented based on multi-feature fusion,which is combined with the gradient,phase feature and edge brightness feature. The algorithm takes advantage of gradient similarity,phase consistency and edge brightness similarity,and describes the local structure features of images; At the same time,the gradient similarity and edge similarity of images can be used to better reflect the contrast and significant changes of images.The objective evaluation method is consistent with the human visual perception by fusing multi-feature similarity. Finally,the experimental results on the TID2008 image database show that the proposed method has better performance for different types of image distortion.