Comfort Evaluation of Stereo Images Based on Machine Learning
Yi Liu, Wei Quan, Cheng Han, Chao Wang, Hua Li, Jiaqi Liu · 2020
Aiming at the difficult problem that how to evaluate comfort of stereo images effectively, this paper proposes a visual comfort evaluation model of stereo images based on machine learning. The job is conducted based on a focus area in images. The disparity map is calculated from a pair of stereo images. A primary salient area is obtained by the GBVS algorithm from right eye image, and the final stereo salient area is generated from both parallax information and primary salient area. Next, multiple types of features in the region are extracted. At last, the support vector regression model is used to predict the visual discomfort score. The experimental result shows that the proposed objective evaluation model is close to the subjective evaluation value, and it has achieved better results.