| 基于色彩感知的无参考图像质量评价 |
| |
| 引用本文： | 李俊峰,方建良,戴文战. 基于色彩感知的无参考图像质量评价[J]. 仪器仪表学报, 2015, 36(2): 339-350 |
| |
| 作者姓名： | 李俊峰 方建良 戴文战 |
| |
| 作者单位： | 浙江理工大学自动化研究所;浙江工商大学信息与电子工程学院 |
| |
| 基金项目： | 国家自然科学基金(61374022);浙江省公益性技术应用研究计划(2014C33109);浙江省新型网络标准及其应用技术重点实验室开放课题(2013E10012)资助项目 |
| |
| 摘 要： | 图像发生失真会改变RGB色彩空间的颜色特征、各色彩分量的亮度分布及其间的相关性,基于此,提出了一种新的无参考图像质量评价方法。首先,标准化6种颜色系数以消除光照环境变化对RGB模型的影响,并利用各颜色系数的拟合广义高斯分布模型(GGD)的形状参数作为颜色统计特征;其次,分别计算各色彩分量的均值减损对比归一化(MSCN)系数及其邻域系数间的互信息,利用互信息作为统计特征来描述其各分量间的相关性;进而,利用人眼更为敏感的G分量MSCN系数的拟合GGD模型参数及其4方向邻域MSCN系数的拟合非对称广义高斯分布模型(AGGD)参数作为亮度统计特征;最后,分别利用支持向量回归机(SVR)和支持向量分类机(SVC)构建无参考图像质量评价模型和图像失真类型识别模型,并在LIVE等数据库上进行了算法与差异平均意见分(DMOS)的相关性、模型的鲁棒性等方面的实验。实验结果表明,本文方法的评价结果与人类主观评价具有高度的一致性;而且图像失真类型识别模型的识别准确率也高达到93.59%,明显高于当今主流无参考图像质量评价方法。
|
| 关 键 词： | 无参考图像质量评价 自然场景统计 色彩感知 互信息 |
No-reference image quality assessment based on color perception |
| |
| Li Junfeng,Fang Jianliang,Dai Wenzhan. No-reference image quality assessment based on color perception[J]. Chinese Journal of Scientific Instrument, 2015, 36(2): 339-350 |
| |
| Authors: | Li Junfeng Fang Jianliang Dai Wenzhan |
| |
| Affiliation: | Li Junfeng;Fang Jianliang;Dai Wenzhan;Institute of Automation,Zhejiang Sci-Tech University;School of Information & Electronic Engineering,Zhejiang Gongshang University; |
| |
| Abstract: | The image distortion can change the color features of the image, the luminance distribution of the color components and the correlation among those components in RGB color space. Based on this, a novel no-reference image quality assessment (NR-IQA) method is proposed. Firstly, six color coefficients are standardized to remove the effect of illumination changes on the RGB model; and the shape parameters of the generalized Gauss distribution (GGD) model fitted from the color coefficients are used as color statistics features. Secondly, the mutual information between the mean subtracted contrast normalized (MSCN) coefficients and the neighboring coefficients of the color components in RGB color space is calculated. The mutual information is taken as the statistics features to describe the correlation among the color components in RGB color space. Moreover, the fitted GGD model parameters of MSCN coefficients and the fitted AGGD model parameters of its four direction neighboring MSCN coefficients of the G component more sensitive to human eyes are taken as the luminance statistical features. At last, combining the above-mentioned statistics features in RGB color space, support vector regression (SVR) and support vector classifier (SVC) are used to construct the NR-IQA model and the image distortion type recognition model respectively. And a large number of simulation experiments were carried out in the LIVE and TID2008 databases in order to analyze the correlation between the algorithm and difference mean opinion score (DMOS), the robustness, the classification accuracy and the computational complexity of the algorithm. The simulation results show that this method is suitable for many common distortions and highly consistent with the human subjective assessment result. In addition, the recognition accuracy of the image distortion type recognition model is up to 93.59%, which is significantly superior to that of the present-day main stream NR-IQA methods. |
| |
| Keywords: | no-reference image quality assessment natural scene statistics color perception mutual information |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《仪器仪表学报》浏览原始摘要信息 |
|
点击此处可从《仪器仪表学报》下载免费的PDF全文 |
|
