Blinder David, Schelkens Peter
Opt Express. 2018 Jan 22;26(2):1461-1473. doi: 10.1364/OE.26.001461.
Computer-generated holography at high resolutions is a computationally intensive task. Efficient algorithms are needed to generate holograms at acceptable speeds, especially for real-time and interactive applications such as holographic displays. We propose a novel technique to generate holograms using a sparse basis representation in the short-time Fourier space combined with a wavefront-recording plane placed in the middle of the 3D object. By computing the point spread functions in the transform domain, we update only a small subset of the precomputed largest-magnitude coefficients to significantly accelerate the algorithm over conventional look-up table methods. We implement the algorithm on a GPU, and report a speedup factor of over 30. We show that this transform is superior over wavelet-based approaches, and show quantitative and qualitative improvements over the state-of-the-art WASABI method; we report accuracy gains of 2dB PSNR, as well improved view preservation.
高分辨率的计算机生成全息术是一项计算密集型任务。需要高效的算法来以可接受的速度生成全息图，特别是对于诸如全息显示之类的实时和交互式应用。我们提出了一种新颖的技术，该技术使用短时傅里叶空间中的稀疏基表示，并结合放置在三维物体中间的波前记录平面来生成全息图。通过在变换域中计算点扩散函数，我们仅更新预先计算的最大幅度系数的一小部分，从而比传统的查找表方法显著加速算法。我们在图形处理器（GPU）上实现了该算法，并报告了超过30倍的加速因子。我们表明，这种变换优于基于小波的方法，并且在定量和定性方面比当前最先进的WASABI方法有所改进；我们报告峰值信噪比（PSNR）提高了2dB，同时视图保留也得到了改善。
