Virtual View Synthesis by a Variational Recovery Method with a Multiple Depth-Layer Model

摘要

This paper presents a novel method for virtual view synthesis using an image recovery scheme based on energy minimization. The presented method first synthesizes multiple virtual views at the same position based on multiple depth layers by using the conventional view interpolation method. The interpolated views suffer from blurring and ghosting artifacts due to the pixel mis-correspondence. Secondly, the multiple views are integrated into a novel view in which all regions are focused. In this paper, we formulate the integration problem as the image recovery problem of minimizing energy that consists of data-fidelity and regularization terms. This integration method effectively recovers the all-focused view, avoiding problems of feature correspondence and scene geometry reconstruction.