基于空域自然场景统计的无参考立体图像质量评价模型

针对现有的评价方法大都将图像变换到不同的坐标域问题,提出一种基于空域自然场景统计(NSS)的通用型无参考立体图像质量评价模型。在评价中为了更好地结合人类双目视觉特性, 将左右图像融合成一幅独眼图;评价模型首先统计独眼图归一化亮度(CMSCN)系数分布规律,进而对独眼图提取空域自然场景统计特征;其次,统计视差图归一化亮度(DMSCN)系数的分布规律,并对用光流法得到的视差图提取同样的特征;最后,通过支持向量回归(SVR)建立立体图像特征信息与主观评价值(DMOS)之间的关系,从而预测得到图像质量的客观评价值。实验结果表明,该评价模型对立体数据测试库进行评价,其Pearson线性相关系数(PLCC)和Spearman等级相关系数(SROCC)值均在0.94以上;对于非对称立体图像库,PLCC和SROCC值分别接近0.91和0.93。该模型能够很好地预测人眼对立体图像的主观感知。

No-reference stereoscopic image quality assessment model based on natural scene statistics

Focusing on the issue that most of the existing evaluation methods transform images into different coordinate domain, a spatial Natural Scene Statistics (NSS) based model of no reference stereoscopic image quality assessment method was proposed. Among the stereoscopic image quality assessment, in order to better combine with the binocular visual features of human beings, left and right images were fused to construct a cyclopean map. Firstly, via statistical distribution of the Cyclopean Mean Subtracted Contrast Normalized (CMSCN) coefficients, the natural scene statistical characteristics were extracted in spatial domain from the cyclopean map. Secondly, by getting statistical distribution of the Disparity Mean Subtracted Contrast Normalized (DMSCN) coefficients, and the same characteristics were extracted from the disparity map obtained by optical flow model. Finally, Support Vector Regression (SVR) was performed to predict the objective scores of stereoscopic images by establishing the relationship between the stereoscopic image feature information and the Difference Mean Opinion Score (DMOS). The experimental results show that compared with other methods, the Pearson Linear Correlation Coefficient (PLCC) and Spearman Rank-Order Correlation Coefficient (SROCC) indicators reach 0.94 on symmetric stereoscopic image database, and the PLCC indicator reaches 0.91 and the SROCC indicator reaches 0.93 on asymmetric stereoscopic image database, which indicate the proposed method can achieve higher consistency with subjective assessment of stereoscopic images.