基于支持向量回归的立体图像客观质量评价模型

摘要

Stereoscopic image quality assessment is an effective way to evaluate the performance of stereoscopic video system. However, how to use human visual characteristics effectively is still a research focus in objective stereoscopic image quality. In this paper, combining with the stability characteristics of singular values and subjective visual characteristics of stereoscopic images, an objective stereoscopic image quality assessment model based on Support Vector Regression (SVR) is proposed. In the model, firstly, stereoscopic features are obtained by extracting singular values of left and right images. Secondly, the features are fused according to different types of distortion. Finally, the values of objective assessment are predicted by SVR. Experimental results show that, by applying the proposed model to stereoscopic test database, Person's correlation coefficient index reaches to 0.93, Ranked correlation coefficient index reaches to 0.94, Root Mean Square Error (RMSE) index approaches to 6, and outlier ratio index reaches to 0.00%, which indicate that the model is fairly good and can predict human visual perception very well.%立体图像质量评价是评价立体视频系统性能的有效途径,而如何利用人类视觉特性对立体图像质量进行有效评价是目前的研究难点.该文根据图像奇异值有较强稳定性的特点,结合立体图像的主观视觉特性,提出了一种基于支持向量回归(Support Vector Regression,SVR)的立体图像客观质量评价模型.该模型通过分析立体图像的视觉特性,提取左右图像的奇异值作为立体图像的特征信息,然后根据立体图像的不同失真类型情况对其特征进行融合,通过SVR预测得到立体图像质量的客观评价值.实验结果表明,采用该文提出的客观评价模型对立体数据测试库进行评价,Pearson线性相关系数值在0.93以上,Spearman等级相关系数值在0.94以上,均方根误差值接近6,异常值比率值为0.00％,符合人眼视觉特性,能够很好地预测人眼对立体图像的主观感知.