| 王勇,王宇庆,顾海军,赵晓辉.客观评价彩色图像质量的全四元数结构相似度方法[J].光电子激光,2014,(10):2033~2043 |
| 客观评价彩色图像质量的全四元数结构相似度方法 |
| Color image quality assessment method based on full quaternion structure similarity measure |
| 投稿时间：2014-02-20 |
| DOI： |
| 中文关键词: 图像质量评价 全四元数 人类视觉系统(HVS) 结构相似度(SSIM ) |
| 英文关键词:image quality assessment full quaternion human visual system(HVS) struc ture similarity (SSIM) |
| 基金项目:国家自然科学基金青年基金(61201368)资助项目 (1、吉林大学 通信工程学院,吉林 长春 130012; 2、中国科学院航空光学成像与测量重点实验室,中国科学院长春光学精密机械与物理研究所,吉林 长春 130033) |
|
| 摘要点击次数: 1577 |
| 全文下载次数: 168 |
| 中文摘要: |
| 在不损失彩色图像颜色信息的基础上,通过定义一 种全四元数模型 将四元数的3个虚部以及实部全部用于表示彩色图像的结构信息,从而在完整保留颜色信息 的基础上提高了四元数矩阵的信息承载量。为了进一步增加与主观感知的一致性,将彩色图 像的细节信息和颜色分量合并作为四元数的实部和虚部,进而改进传统的图像质 量评价结构相似度(SSIM)方法,采用四元数结构相似度方法 实现了对待测图像的质量评价。实验结 果表明,本文方法既保留了彩色图像的颜色信息,又增强了图像内容中人眼 敏感的细节信息,采用LIVE数据库和IVC数据库验证了本文方法的评价指标,实验结果 表明, 所提指标与人眼主观感知的一致程度要好于传统的图像质量评价方法以及纯四元数SSIM方法。在交叉失真实验中,量化评价结果的Kendall等级相关系数 (KRCC),Spearman等级相关系 数(SRCC)以及均方根误差(RMSE)为0.6929、0.881以及 7.506,均好于实验中采用的其它几种方法。 |
| 英文摘要: |
| Color information is ignored in most of the image quality assessment m ethods.Color information and luminance information are two important parts in color image.Loss of color info rmation will result in the low consistency of the assessment results.Aiming to the problem,full-quaternion is defined in this paper.It is used to describe the structure information of a col or image.The real part and imagery parts in full quaternion are used to perform the task.It differs from the tradi tional pure quaternion-based color image description method where the detail information in an image is enha nced by using the real part of a quaternion.To enhance the detail information in an image,local variance of lum inance layer and the three color channels are taken as the real part and imagery parts respectively.On the basis of that,the structure similarity (SSIM) measure is studied and the quaternion-based structure similarity measure is use d to perform color image structure comparison.Results from the experiments show that the proposed method combines detail information and color information of a color image.It can describe the structure information better t han the traditional description methods.The quantitative results reveal that the performance of the full quatern ion-based method is better than the pure quaternion-based method and many state-of-the-art methods.The values of Kendall rank correlation coefficient (KRCC),spearman rank correlation coeffici ent (SRCC) and root mean squared error (RMSE) are 0.6929,0.8818and 7.506,respectively,which proves that t he performance of the proposed method is better than the other methods in the experiment. |
| 查看全文 下载PDF阅读器 |
| 关闭 |
|
|
|
|
|