基于视觉重要区域的立体图像视觉舒适度客观评价方法

摘要

Stereo video technology provides depth perception and immersion visual experience, but it also makes people feel visual fatigue and causes the decrease of the experience quality. Therefore, how to evaluate the visual comfort of stereoscopic image/video effectively is still a research focus. In this paper, an objective visual comfort assessment metric based on visual important regions is proposed. First, the Visual Important Regions (VIR) are obtained from image saliency and disparity map information. Then, disparity amplitude, disparity gradient, and spatial frequency features are extracted and fused into a feature vector. Finally, the values of objective assessment are predicted by Support Vector Regression (SVR). Experimental results show that compared with existing methods, the proposed metric achieves higher consistency with subjective visual comfort assessment of stereoscopic images.%先进的立体视频技术能给观众带来深度感和沉浸感，但也容易使人产生视觉疲劳，造成观看体验质量的下降，因此，如何对立体视频/图像的视觉舒适度进行有效评价是目前的研究难点。该文提出一种基于视觉重要区域的立体图像视觉舒适度客观评价模型，该模型主要包括3个部分：(1)利用图像显著图和视差图像得到立体图像视觉重要区域；(2)提取视觉重要区域的视差幅度特征、视差梯度边缘特征以及空间频率特征，作为反映立体图像视觉舒适度的感知特征信息；(3)通过支持向量回归建立特征信息与立体图像舒适度平均主观评分值的关系，预测得到立体图像视觉舒适度的客观评价值。实验结果表明，与现有的方法相比较，在相同的立体图像测试库上，采用该文提出的客观评价模型可以获得更好的评价性能。