3DVSR: 3D EPI volume-based approach for angular and spatial light field image super-resolution

Light field (LF) imaging, which captures both spatial and angular information of a scene, is undoubtedly beneficial to numerous applications. Although various techniques have been proposed for LF acquisition, achieving angularly spatially high-resolution remains technology challenge. In this paper, learning-based approach applied 3D epipolar image (EPI) reconstruct LF. Through 2-stage super-resolution framework, the effectively addresses (SR) problems, i.e., SR, angular-spatial SR. While first stage provides flexible options up-sample EPI volume desired resolution, second stage, consists novel volume-based refinement network (EVRN), substantially enhances quality volume. An extensive evaluation on 90 challenging synthetic real-world light scenes from 7 published datasets shows that outperforms state-of-the-art methods large extend problem, an average peak signal noise ratio improvement more than 2.0 dB, 1.4 3.14 dB in SR ×2, ×4, respectively. The reconstructed 4D demonstrates balanced performance distribution across all perspective images presents superior visual compared previous works.