Abstract: The authors present a method to detect the stereo lens distortions, such as vertical misalignment, camera rotation, unsynchronized zooming, and color misalignment, which are introduced during the process of video capturing. Based on the result of stereoscopic correspondence, the geometrical errors are calculated for every matched feature point pairs. The overall lens distortion parameters are deduced from the statistical features of all points. Experiments are set on a video database which consists of video clips from an on-air 3DTV channel. Experiment result shows that the method is effective to find out distorted video clips. The subjective quality assessment experiment reveals some relationships between the lens distortions and the perceptual video quality.

Key words: stereoscopic video, lens distortion, 3DTV, video quality

摘要： 针对立体视频在拍摄过程中由于镜头配置和调整导致的立体效果的失真, 如左右眼图像的高度误差、旋转误差、尺寸误差、亮度色度误差等, 提出一种立体镜头误差检测的方法。在对左右眼图像进行立体匹配的基础上, 根据图像的几何关系计算每对匹配特征点的各种误差值, 然后计算所有误差值的统计特征, 得到图像的总体误差值。在收录的实际拍摄的3DTV电视节目视频库中进行的实验表明, 该方法能够有效识别存在立体镜头误差的视频片段。进一步的主观视频质量实验分析了立体镜头误差与主观质量的关系。

关键词: 立体视频, 镜头失真, 3DTV, 视频质量