Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (2): 154-161.DOI: 10.3778/j.issn.1002-8331.1709-0436

Previous Articles     Next Articles

No-Reference Image Quality Assessment Algorithm for Stereoscopic Images via Dual-Tree Complex Wavelet Transform

GU Tingting, LIU Xinhui, SANG Qingbing, LI Chaofeng   

  1. School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2019-01-15 Published:2019-01-15

基于双树复小波的无参考立体图像质量评价

顾婷婷,刘新会,桑庆兵,李朝锋   

  1. 江南大学 物联网工程学院,江苏 无锡 214122

Abstract: With the rapid development of 3D technology, stereoscopic images are widely used in many fields. At the same time, it is expected that high-definition images should be provided. So image quality assessment for stereoscopic images has been the concern and a no-reference image quality assessment algorithm for stereoscopic images via Dual-Tree Complex Wavelet Transform(DT-CWT) is proposed. Firstly, the left and the right view images of a stereoscopic image need to be processed in the way of DT-CWT to generate the textural structure image. And the parallax image is obtained according to the principle of minimum energy error. Secondly, features from the textural structure image and the parallax image are extracted, including the parameters of Asymmetric Generalized Gaussian Distribution(AGGD), variance of Gradient Magnitude(GM) and Relative gradient Orientation(RO), the area of Singular Value Decomposition(SVD) curve and the coordinate axis. Lastly, it trains AdaBoosting Back-Propagation(BP) neural network by utilizing these features and predicts the quality scores of stereoscopic images. The experiment on LIVE 3D Image Quality Database demonstrates that the predicted scores by the new method have high correlation with mean opinion scores and a good evaluation result is obtained.

Key words: dual-tree complex wavelet transform, asymmetric generalized Gaussian distribution, gradient magnitude, relative gradient orientation, singular value decomposition, AdaBoosting back-propagation neural network

摘要: 随着3D技术的不断发展,立体图像的使用领域越来越广泛,同时人们对图像的清晰度要求越来越高,因此,立体图像的质量评价成为关注点,基于此,提出了一种基于双树复小波变换的立体图像质量评价算法。使用双树复小波变换对立体图像的左、右视图进行处理,生成纹理结构图像,且根据最小能量误差的原理,获取左右视图的视差图;对纹理结构图像和视差图提取非对称广义高斯分布模型的参数、梯度幅值、相对梯度方向方差和奇异值曲线与坐标轴的面积等特征;使用AdaBoosting BP神经网络,进行训练和预测立体图像的质量得分。在LIVE立体图像数据库上的实验结果表明,新方法预测得分与主观得分有较好的一致性,获得了比较好的实验结果。

关键词: 双树复小波变换, 非对称广义高斯分布, 梯度幅值, 相对梯度方向, 奇异值, AdaBoosting BP神经网络