基于空间依存的无参考图像质量评价

为了实时监测图像质量,建立了像素小波系数的二元空间依存关系模型,并利用该模型实现了图像质量的无参考评价。首先,将RGB图像映射到HSV空间;对图像进行小波分解,并建立小波系数的二元空间依存关系模型,即以广义高斯分布来拟合小波系数的二元联合分布。然后,分析二元空间依存关系与图像质量的相关性,建立了无参考图像质量评价指标。最后,对图像质量评价指标进行了测试及对比研究。基于TID2013、LIVE及CSIQ数据库完成了测试,结果表明:基于空间依存的无参考图像质量评价指标可以对图像的失真程度进行准确分级,分级准确率达到96%以上;采用基于空间依存的无参考图像质量评价方法可以实现对图像质量失真度的准确分级。

No-reference image quality assessment based on spatial dependency

To monitor the image quality in real-time, a binary spatial dependence model of pixel wavelet coefficients was established, and the model was used to realize the image quality assessment by the no-reference image method. Firstly, the RGB image was mapped into a HSV(Hue,Saturation,Value) space and was processed by wavelet decomposition. A binary space dependent relationship model of wavelet coefficients was established, in which the generalized Gaussian distribution was used to fit the binary joint distribution of wavelet coefficients. Then, the correlation between the binary spatial interdependence relationship and the image quality was analyzed, and the no-reference image quality assesment index was obtained. Finally, the proposed image quality assessment indexes were studied and tested comparatively based on the TID2013, LIVE and CSIQ databases. The results show that the image quality assesment index based on the spatial dependency can be used to classify the image distortion degree accurately, and the classification accuracy rate reaches above 96%. It concludes that proposed no-reference image method based on the spatial dependency achieves accurate image quality classification.