基于扩展梯度算子的结构相似度图像质量评价方法

为了更精确地评估图像质量,提出了一种新的客观评价算法:基于扩展梯度算子的结构相似度图像质量评价方法(extended gradient-based structural similarity,E-GSSIM)。首先分析了结构相似度(structural similarity,SSIM)与梯度结构相似度(gradient-based structural similarity,GSSIM)的不足,提出应用扩展Sobel算子替代GSSIM的传统Sobel算子,从而能够从多个方向提取邻域的梯度信息。为了不破坏原图的固有图像属性及使所提取的梯度信息更具有一致性,在计算梯度信息的时候引入衰减与阈值因子。LIVE2、TID2008、TID2013与CSIQ四个图库的定性与定量验证表明,E-GSSIM算法要优于传统的PSNR、SSIM、GSSIM质量评价指标,更为符合人眼视觉感知结果。

An Image Quality Assessment Metric based on Structure Similarity of Extended Gradients

In order to evaluate image quality more accurately, a novel objective evaluation algorithm is proposed: Extended Gradient-based Structural Similarity (E-GSSIM). Firstly, the disadvantages of Structural Similarity (SSIM) and Gradient-based Structural Similarity (GSSIM) are analyzed in this paper. Then the extended Sobel operator is employed to replace the traditional Sobel operator in the computing procedure of GSSIM so that the gradients in an image can be extracted from multiple directions. In order to avoid destroying the inherent attributes of the original image and to let the extracted gradient information be more consistent, an attenuation factor and a threshold are employed in this paper. The qualitative and quantitative experimental results based on four image datasets of LIVE2, TID2008, TID2013 and CSIQ show that the E-GSSIM outperforms the traditional PSNR, SSIM, and GSSIM, and is more consistent with the human visual perceptions.