结合立体视频帧左右视图的灰度图; 采用鲁棒性算法与软剪刀算法相结合的“画笔”方案来提取立体视频帧的前景区域; 合成视差图并提取水平视差; 计算深度视差; 根据主观评价结果建立视差深度-视觉舒适度模型; 再结合亮度对视觉舒适度的评价模型进行优化; 使模型的评分更接近人类视觉系统的评分; 结合人类主观评价实验的结果; 用逆向代入法求出模型的系数确定模型; 最终实现基于人类视觉系统特性的对立体视频视觉舒适度的评价。实验结果表明：通过视觉舒适度模型计算得到的结果与人类主观评价结果之间的误差率不超过5%; 其中大部分视频误差率不超过1%。说明评价模型更符合人类主观感知结果的立体视频视觉舒适度; 模型评分更接近人类视觉系统特性的主观实验评分; 这为立体视频舒适度的评价提供了参考。

We extract the foreground region of stereoscopic video by combining the grayscales of left and right view, and synthesize the two pictures into a disparity map, then extract the horizontal parallax and calculate the depth parallax. Finally; we build the depth parallax-visual discomfort model on the basis of subject assessment and optimize the visual discomfort model by considering the luminance influence. Experimental results indicate that the error rate is less than 5% between calculated value and human subjective evaluation results, and most video error rates are less than 1%. The evaluation model is more accord with human subjective perception results and has good practicability, which provides a reference for the stereo video comfort evaluation.