Learning from EPI-Volume-Stack for Light Field image angular super-resolution

作者：

Highlights：

• A light field image angular super-resolution network based on light field EPI-Volume-Stack representation is proposed.

• A new light field EPI-Volume-Stack representation for light field image angular super-resolution is introduced to explore light field view spatial and angular correlations.

• A learning based light field method is proposed, in which 3D convolutional operations are adopted to better accommodate EPIVolume-Stack data and allow information propagation between two spatial and one directional dimensions of EPI-Volume-Stack data.

• Extensive experiments on real world scenes, synthetic scenes, and LF application validate the effectiveness of the proposed method in achieving a better super-resolution quality and restoring more texture details of synthesized light field views.

摘要

•A light field image angular super-resolution network based on light field EPI-Volume-Stack representation is proposed.•A new light field EPI-Volume-Stack representation for light field image angular super-resolution is introduced to explore light field view spatial and angular correlations.•A learning based light field method is proposed, in which 3D convolutional operations are adopted to better accommodate EPIVolume-Stack data and allow information propagation between two spatial and one directional dimensions of EPI-Volume-Stack data.•Extensive experiments on real world scenes, synthetic scenes, and LF application validate the effectiveness of the proposed method in achieving a better super-resolution quality and restoring more texture details of synthesized light field views.

论文关键词：Light field image angular super-resolution,EPI-volume-stack,3D convolution,Deep learning

论文评审过程：Received 7 October 2020, Revised 22 April 2021, Accepted 7 June 2021, Available online 12 June 2021, Version of Record 16 June 2021.

论文官网地址：https://doi.org/10.1016/j.image.2021.116353