尺度参数分布斜率模型的无参考图像质量评价算法

No-reference Image Quality Assessment via Slope Distribution of Scale Parameters

无参考图像质量评价是一种无需任何先验知识的图像质量评价算法.大多数无参考评价算法基本都是针对某种类型的失真而不能较好地应用于其他类型.因此,设计一种能够广泛应用的无参考评价模型非常必要.文章采用小波变换分析图像失真特点,利用广义高斯分布对小波系数分布进行拟合,并分析各尺度参数分布的斜率特性,建立测量失真程度模型,提出了一种基于斜率分布的无参考图像质量评价方法.该算法利用LIVEⅡ图像质量评价数据库进行仿真测试.实验结果表明,算法不仅能够适应大部分失真类型,而且图像质量评价结果与主观评价结果有较好的一致性.

No-Reference image quality assessment can predict image quality without the information of original image. Most No-Reference image quality assessments which have been already proposed are designed for one or a set of specific distortion types and are unlikely to evaluate images of other types of distortion. So there is a strong need to build a No-Reference image quality assessment method which can be widely applied. We decompose the test image by wavelet and model the marginal distribution of coefficients by generalized Gaussian density model, and evaluate image quality by the character of slope distribution. Our algorithm is tested on the LIVE database Ⅱ, experimental results demonstrate that our method can be widely used in most distortions, and obtains good quality of prediction performance.