光场由于可承载相对于图像更完整的场景信息而具有广阔的研究价值和应用前景。由于其数据上的高维属性，直接采集密集采样光场在实际应用中可行性较低，因此利用少量高分辨率视图重构密集采样光场是当前研究热点之一。光场重构问题有稀疏角度采样和非朗伯两大挑战。此外，允许灵活的输入位置和灵活的输出角度分辨率也是必要的。为解决这三个问题，本项目在稀疏重构框架下，提出几何感知的光场自适应表示与混合重构框架，具体包括：（1）建立场景的多级局部非均匀分层表示，为场景几何结构近似和表示提供有力工具；（2）构造适用于光场稀疏表示的多维非均匀滤波器组，在几何信息指导下实现光场的自适应稀疏表示；（3）构造基于稀疏正则化—深度学习的混合重构框架，实现光场的稳定可控重构。本项目以满足诸多实际应用对光场的需要为目标，旨在给出一个灵活、高效、精确的密集采样光场重构解决方案，以期实现光场重构及相关逆问题的理论突破和技术创新。

The light field has great research value and application prospect because it can carry more complete scene information than image. Due to the high dimensionality nature of the data, it is less feasible to directly capture densely sampled light fields in practical applications. Therefore, reconstructing densely sampled light fields using a small set of views with high spatial resolution is one of the current research hotspots. Light field reconstruction faces two major challenges: angular sparsity and non-Lambertian. Moreover, it is necessary to enable flexible input positions as well as flexible output angular resolution. To solve these problems, this project this project proposes a geometry-aware adaptive light field representation and hybrid reconstruction framework under the sparse reconstruction framework. The detailed research proposal includes: (1) build the multi-level local nonuniform layered representation of the scene, which provides a powerful tool for scene geometry approximation and representation; (2) construct the multi-dimensional nonuniform filter bank for sparse representation of light field, whose subband partitioning is guided by the scene a priori geometric information, to realize adaptive sparse representation of light field; (3) establish the hybrid reconstruction framework based on sparse regularization-deep learning, to finally enable a stable and controllable reconstruction from undersampled light field. This project aims to meet the needs of various practical applications for light field, and will provide a flexible, efficient and accurate solution for densely sampled light field reconstruction, in order to achieve theoretical breakthrough and technological innovation of light field reconstruction and related inverse problems.