Learning based no-reference metric for assessing quality of experience of stereoscopic images

Abstract

Human’s perception plays a very important role on image assessment, especially for stereoscopic images. In general, viewing stereoscopic 3D images will cause visual fatigue, eyestrain, dizziness or headache. Therefore, how to evaluate human’s perception of visual quality on 3D images becomes an emerging topic. In this paper, we propose a no-reference assessment metric for stereoscopic image quality of experience (QoE). First, the stereoscopic image pairs are used to calculate the disparity maps by optical flow estimation. Then the depth information are extracted from the disparity map, called as disparity-depth map. Next, we extract four types of features based on pixel value and distribution of disparity-depth map. Two regression models are used to predict visual discomfort scores. Also, we test the proposed method on EPFL 3D image database and IEEE-SA stereoscopic image database, respectively. The experiment results show that our proposed QoE assessment metric achieves excellent performance compared with state-of-the-art methods.