基于非负矩阵分解的光场显示方法

Light field theory directly models the distribution and propagation of lights in the space. It avoids the complexity of traditional geometric modeling, and hence has its advantages in virtual reality, computational photography and other applications. The methods for light field representation and capturing have been relatively mature, but the light field displaying technology is not yet perfect. Aiming to solve the displaying problems, in this project we will study on the light field displaying theory and applications, and propose a novel and more practical lightweight displaying system on the basis of existing technology...In our displaying system, a multi-layered LCD will be used for autostereoscopic display of given light fields. The images displayed on each layer of LCD are optimized according to the target light field by an algorithm based on non-negative matrix factorization (NMF). The overlapped images modulate the lights enter the observer's eyes to form the desired light field...In order to achieve high displaying resolution and wide observing range, we will utilize the features of the frequency domain of the light field to reduce the computational complexity of the image optimization. And the optimization problem for a wide observing range can also be turned in to optimization problems for a series of small ranges, which are much easier to solve. Then the optimized sub-light-fields of the subdivided observing ranges can be displayed by time division multiplexing. We will also explore the parallelizability of the displaying method, and accelerate it by a multi-GPU-architecture.

光场是直接对空间中光线的分布和传播进行建模，避免了传统几何建模的复杂性，在虚拟现实、计算摄影学等领域中都具有其优势。对光场进行表达和采集的方法已相对成熟，但光场显示技术尚不完善。针对这一问题，本项目将对光场显示的理论和应用进行相关研究，在现有技术基础上提出一种新的更易于实际应用的小型光场显示系统。.该系统中将采用多层液晶显示屏来实现裸眼光场自动立体显示。其中在每层液晶屏上显示的图像根据目标光场通过基于非负矩阵分解的算法进行优化，从而使多层图像叠加调制后形成进入人眼的子光场。为了提高显示分辨率、扩大光场可视范围，我们将利用光场的频域特性，减少光场图像优化的计算量。同时通过时分复用策略将较大可视范围的优化问题转化为多个小范围优化问题，从而降低复杂度，提高算法可并行性。我们还将进一步利用多GPU并行计算框架对该光场显示方法进行硬件加速，以实现光场的实时绘制。