基于四元数的彩色图像质量评价方法  

Assessment Method for Color Image Quality Based on Quaternion

作　　者：王宇庆[1,2] 刘维亚[1] 王勇[1,2] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033 [2]中国科学院研究生院,北京100039

出　　处：《中北大学学报（自然科学版）》2010年第1期59-64,共6页Journal of North University of China(Natural Science Edition)

基　　金：国家科技创新基金资助项目(020262112);吉林省与中国科学院科技合作资金资助项目(2005SYHZ0010);中国科学院东北振兴科技行动计划重点资助项目(DBZX-2-017)

摘　　要：提出了一种结合人眼视觉特性与颜色信息的彩色图像质量评价方法.将四元数用于表征彩色图像的结构信息,四元数的3个虚部分别表示彩色图像的三基色分量.为了突出图像的细节信息对图像质量的影响,将彩色图像亮度分量的局部方差作为四元数的实部,从而得到了包含彩色图像结构信息的四元数矩阵.将四元数运算转换为复数运算,求得四元数矩阵的奇异值特征向量,通过计算两奇异值特征向量夹角的大小度量二者所表征的待测图像与参考图像的结构相似程度,从而实现了对待测图像的质量评价.实验结果表明:该方法的评价结果优于传统的MSE,PSNR以及SSIM方法,与人的主观感觉的一致性较好,并且也可以用于评价与参考图像尺寸不相等的待测图像的质量.Combination of color information and properties of human visual system was used to assess the quality of color images. Structural information of a color image was described by quaternion. The three RGB channels of a color image were encoded into the three imaginary parts of a quaternion. The real part of a quaternion was the local variance of the luminance layer of the color image and the effect of the detail structure on the image quality was emphasized accordingly. An algorithm to calculate the singular value decomposition （SVD） of a quaternion matrix using its equivalent complex matrix was used to perform the decomposition of the image. The angle between singular vectors of the reference image and the distorted image was used to measure the structural similarity of the two images. Result from experiments shows that the proposed method is better consistency with human visual system characteristics than MSE, PSNR and SSIM. It can also be used to assess the images whose size is different from that of the reference image.

关 键 词：图像质量评价 四元数 局部方差 奇异值分解 

分 类 号：TN911.73[电子电信—通信与信息系统]