基于非下采样轮廓波变换和多核学习的盲图像质量评价

Blind image quality assessment based on non-subsampled contourlet transform and multiple kernel learning

非下采样轮廓波(Contourlet)变换具有多尺度、多方向特性,能够对图像纹理和结构信息进行精确提取,可以很好地模拟人类视觉系统的多分辨率特性,基于此提出一种基于非下采样Contourlet变换的通用型盲(无参考)图像质量评价算法。首先在空间域上对图像进行非下采样Contourlet变换;然后在各方向带中分别提取能有效反映人类视觉失真程度的特征:高频幅值、平均梯度、信息熵作为图像的特征;最后将其输入到高效的分层多核学习机中学习,预测图像的质量得分。在混合失真型数据库和3个单失真型数据库上的交叉实验结果表明,该算法性能优越,能很好地预测失真图像质量,具有很好的主客观一致性。

The non-subsampled contourlet transform has multi-scale and multi-directional character- istics, which can extract the image texture and structure information accurately an the multi-resolution characteristic of the human visual system. Based on this, we propose a blind image quality assessment algorithm based on non-subsampled contourlet transform. Firstly, the algorithm de- composes the images on spatial domain by non-subsampled contourlet transform. Secondly, the features such as high frequency amplitude, average gradient and information entropy, which can effectively re- fleet the characteristics of human visual distortion degree, are extracted in each direction. Finally, the features are input into the efficient multi kernel learning machine to learn and predict image quality scores. Cross experimental results on multi-kind distortion database and three single distortion databases show that the algorithm is superior in performance and can predict image quality distortion well and has very good subjective and objective consistency.