分析了传统的基于全局视差特性的像视觉舒适度评价模型的不足, 提出了一种基于显著性分析的立体图像视觉舒适度客观预测模型。首先, 根据人眼的立体视觉注意力机制, 利用协方差矩阵和Sigma特征集分别计算得到图像显著图和深度显著图, 并组合得到立体显著图; 然后, 利用立体显著图加权得到立体图像视觉舒适度感知特征; 最后, 通过支持向量回归构造视觉舒适度预测函数, 建立视觉舒适度感知特征和主观评价值之间的关系模型, 从而预测得到立体图像视觉舒适度客观评价值。实验结果表明, 本文评价方法的Pearson线性相关系数值达到0.79, Spearman等级相关系数值达到0.81, 表明提出的模型更加符合人眼视觉特性, 得到的客观评价值与主观感知具有较高的一致性。

The drawbacks of the traditional visual comfort assessment metrics for stereoscopic images by using only global disparity features were analyzed. An objective visual discomfort prediction model of stereoscopic images was proposed based on visual saliency analysis. Firstly, an image saliency map and a depth saliency map were calculated by using covariance matrices and Sigma feature sets respectively according to the stereo visual attention mechanism of human eyes and the stereoscopic saliency map was obtained by combination of the two calculations. Then, visual discomfort perceptual features were obtained by using the stereoscopic saliency map as weighting. Finally, the relationship between the visual discomfort perceptual features and the subjective scores was established by constructing a visual discomfort prediction function with support-vector regression, and the objective visual comfort scores were predicted. Experimental results show that the Pearson Linear Correlation Coefficient (PLCC) index of the proposed method reaches 0.79, and the Spearman Rank Order Correlation Coefficient (SRCC) index reaches 0.81. These results indicate that the proposed model can achieve higher consistency with subjective perceptual of stereoscopic images, and is more consistent with human visual systems.