Toward fully automatic channel-mismatch detection and discomfort prediction for S3D video

Channel mismatch - i.e., swapped left and right views - can cause major visual discomfort when viewing stereoscopic content, but detecting it automatically remains a challenge. We present a novel method for detecting channel mismatch that significantly outperforms a prior approach when analyzing a dataset of 1,000 video clips. Moreover, this method enabled detection of 65 scenes exhibiting channel mismatch in 105 real S3D movies. Perceived discomfort can vary greatly between different scenes with channel mismatch depending on a number of factors. We conducted a study involving 59 participants to compute subjective discomfort scores for 56 scenes with channel mismatch. In this paper, we propose a model to predict these scores on the basis of scene characteristics.