A Novel Scheme for View Interpolation from Multi-View Images without Disclosure of Entire 3D Information

Abstract

We propose a novel method of high-quality view interpolation without disclosing entire 3D information like DIBR based on prediction using stereo matching. In this paper, by effectively embedding disparity information only on unpredictable regions in multi-view images as frequency components, our method maintains compatibility with applications using no 3D information.