Objective measurement of stereoscopic 3D content-induced visual discomfort using facial expression recognition

Abstract

It is generally agreed that a possible obstacle to the widespread deployment of 3D services is related to issues of visual comfort. To address the issues of viewing safety in stereoscopic 3D television, it is necessary to develop a way to measure the visual discomfort. However, conventional discomfort measurement methods (e.g., subjective discomfort measurement, ocular response measurement, brain activity measurement, etc) have a limit to directly apply real applications due to several practical issues such as expensive equipment, intrusive use of the equipment, etc. This thesis aims at developing a novel discomfort measurement method by analyzing viewers’ facial expression. In order to understand the visual discomfort by stereoscopic viewing, this thesis investigates determinants of visual discomfort based on the relationship between stimulus characteristics and binocular fusion limit. In addition, we investigate the combined effect of individual discomfort causes on the overall visual discomfort. With the understanding of visual discomfort in stereoscopic displays, this thesis further investigates human facial expressions associated with visual discomfort from a face captured by a camera. Based on our analysis, this thesis introduces a new discomfort-measurement method for the analysis of faces viewing stereoscopic display. Experimental results showed that the proposed method was feasible for challenging face videos with fully spontaneous expressions and occlusions via passive 3D glasses (classification accuracy was 82.08%). It is worth to note that the objective discomfort assessment using facial expression analysis could have advantages such as being low cost, of non-intrusive use, and easy in terms of implementation compared to pre-existing objective discomfort measurement tools such as fMRI, EEG, etc.