针对传统的图像质量评价方法忽略颜色信息以及与人眼感知信息一致性差的问题, 提出了一种全面利用彩色图像颜色信息, 突出人眼敏感图像结构的彩色图像质量客观评价方法。将图像中人眼敏感的结构分为细节, 亮度和颜色3方面因素, 将四元数矩阵作为载体, 构造了一种用于彩色图像质量评价的四元数矩阵, 并对其进行奇异值分解。将最大奇异值作为度量图像结构相似性的主要参数, 通过分析图像结构差异映射图谱得到了最终的量化评价结果。采用LIVE数据库中包含5种失真类型的982张测试图片验证了提出的算法, 得到的交叉失真实验非线性拟合均方根误差(RMSE)值为9.176, Spearman等级相关系数(SROCC)值为0.929 6, 而结构相似度(SSIM)方法的RMSE值为9.299, SROCC值为0.925 6。试验结果表明, 该方法采用四元数矩阵描述彩色图像的结构信息, 考虑了彩色图像的多方面结构特征, 与人眼视觉感知特性的一致性优于传统方法。

A new color image assessment method is proposed to solve the problem of neglecting color information and a poor consistent behavior with the human visual system in traditional image quality assessment methods. The human eye sensitive image structure is enhanced and full color information is used in this method. Three important parts are taken into account in this structure, which are detail information, luminance information and color information. Quaternion is taken as a tool to perform the task. A quaternion matrix is constructed to evaluate color image quality. Then singular value decomposition is performed on the quaternion matrix. Max singular value is used to describe image structure information. Numerical results are obtained by using distortion map. 982 images in LIVE database including five types of distortion are used to investigate the behaviors of the proposed method. The nonlinearity property of the proposed method in the cross-distortion experiment is that the Root Mean Square Error(RMSE) value is 9.176,and Spearman Rank Order Correlation Coefficient(SROCC) value is 0.929 6, however, those of from Structural Similarity Index( SSIM) method are 9.299 and 0.925 6, respectively. The results show that proposed method is more consistent than the traditional methods because of the considering of more image properties and using a quaternion matrix.