Abstract:Due to hardware restriction, it is costly to capture densely-sampled Light Fields (LFs) with high angular and spatial resolution, which becomes the main bottleneck of LFs development. In this paper, we propose a learning-based framework to synthesize novel views and reconstruct densely-sampled LFs from sparsely-sampled LFs. In the proposed framework, micro-lens image stacks and view image stacks are separately grouped, in which details in novel views are explored from spatial and angular domain. The two kinds of stacks contain epipolar information and 3D convolution layers are employed to effectively extract features that include structure information. Moreover, an innovative way is proposed to synthesize views by upsampling micro-lens image stacks using deconvolution layers. The parameters in decovolution layers provide the view position information and different interpolation and extrapolation tasks can be explicitly modeled. It is validated that this view synthesis module can be embedded in different frameworks and improve the related performances. Without precise depth estimation and view warping, the proposed method is mainly designed for reconstructing LFs with small baselines. Related experimental results show that the proposed model outperforms other state-of-the-art methods in terms of both visual and numerical evaluations. Furthermore, the consistency between synthesized views and the intrinsic structure information is well preserved in the proposed method.

Multiview from Micro-Lens Image of Multi-Focused Plenoptic Camera

Three Dimensional Reconstruction Using a Lenslet Light Field Camera

Depth-of-Field-Extended Plenoptic Camera Based on Tunable Multi-Focus Liquid-Crystal Microlens Array

Light field camera all-in-focus image acquisition based on angular information

Simultaneous Multiview Capture and Fusion Improves Spatial Resolution in Wide-Field and Light-Sheet Microscopy

Rapid Image Deconvolution and Multiview Fusion for Optical Microscopy.

A Wide Field-of-View Light-Field Camera with Adjustable Multiplicity for Practical Applications

Chromatically Multi-Focal Optics Based on Micro-Lens Array Design.

Geometry Parameter Calibration for Focused Plenoptic Cameras.

Point Spread Function for the Wide-Field-of-view Plenoptic Cameras

Perspective-consistent multifocus multiview 3D reconstruction of small objects

Multi-view stereo using multi-luminance images

Blur aware metric depth estimation with multi-focus plenoptic cameras

Multi-Focus Image Fusion via PAPCNN and Fractal Dimension in NSST Domain

Depth Profile Measurement with Lenslet Images of the Plenoptic Camera

Super-resolved Refocusing with a Plenoptic Camera

Fully Convolutional Network-Based Multifocus Image Fusion.

Micro-Lens Image Stack Upsampling for Densely-Sampled Light Field Reconstruction

Hybrid Light Field Imaging for Improved Spatial Resolution and Depth Range

Image formation analysis and light field information reconstruction for plenoptic camera 2.0

Fractal Dimension-Based Multi-Focus Image Fusion via Coupled Neural P Systems in NSCT Domain